fix(web): usable past-conversation list, and search that finds past sessions

Two home-screen reports from @jordan8037310, both about history that is
present but unreachable.

#260 — "Resume Conversation" rendered 4 rows, then a button that appended
every remaining row into a `max-height: 240px` box, so 35 conversations
landed in a four-row scroll well with no ordering or filtering. Rendering
now goes through `_renderHistoryList()` over a cached corpus: 10 rows to
start, Show more/Show less that grows and shrinks the box (the height cap
is class-driven, `.history-list.expanded`), plus a filter box (name,
folder, #case label, prompts), a sort control (recent / name / folder,
pinned rows still first) and a shown-of-total count. A filter implies
expansion, so every match is visible, and the whole header hides as one
unit while a federated search is active. The A-Z sort keys off the same
string the row renders, since most rows are transcript-backed and carry
no session name at all.

#261 — the search box could not match a past project by folder name:
`harvestSources()` built its session corpus from the live in-memory map,
while past sessions come from `/api/sessions/unified` (lifecycle log +
transcript scan). Folding that scan into the request path would have cost
the search its no-filesystem-reads property, so the corpus arrives via a
bounded snapshot instead: `session-history-index.ts` is published as a
side effect of `/api/sessions/unified` (the home screen fetches it on
open, which is the same screen the search box lives on) and rebuilt
fire-and-forget, single-flight and TTL-guarded when a search finds it
stale. A result for a closed session now resumes the conversation rather
than selecting a tab that no longer exists, and is badged RESUME.

The snapshot is stored unscoped with a per-row owner and re-filtered
through canAccessOwned() on read, so multi-user sees exactly what
/api/sessions/unified exposes: own sessions only, host-wide transcript
history admin-only. Live rows are harvested first and win the dedupe.

Verified end-to-end against a real instance with 60 past sessions: cold
process answers its first search without history and its second with it;
folder-name queries return resume targets; clicking one posts the right
resumeSessionId + workingDir.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

src/web/session-history-index.ts

@@ -0,0 +1,166 @@
+/**
+ * @fileoverview Bounded in-memory index of PAST sessions, harvested by `GET /api/search`.
+ *
+ * `GET /api/search` used to build its session corpus from the live in-memory
+ * session map alone, so a folder sitting in the home screen's "Resume
+ * Conversation" list matched nothing (issue #261). The corpus that list renders
+ * comes from `GET /api/sessions/unified`, which reads the lifecycle log and every
+ * Claude transcript file — disk I/O the search path deliberately does not do
+ * (its no-fs property is what keeps a per-keystroke query cheap and traversal-free).
+ *
+ * This module is the seam between the two: a capped snapshot of the unified list
+ * that the search route reads synchronously, refreshed OUT of the request path.
+ * Two things fill it:
+ *   1. `/api/sessions/unified` writes it as a side effect (free — it just merged
+ *      that list). The home screen calls that endpoint whenever it opens, which
+ *      is the same screen the search box lives on, so it is warm in practice.
+ *   2. `ensureHistorySessionIndexFresh()` — fire-and-forget, single-flight,
+ *      TTL-guarded — kicks the registered refresher when a search finds the
+ *      snapshot stale. The caller never awaits it: the current query answers from
+ *      the existing snapshot and the next one sees fresh data.
+ *
+ * OWNERSHIP: each item carries the `owner` of the session it came from, and rows
+ * not tied to any live/persisted session (host-wide transcript history) carry
+ * `owner: undefined`. `canAccessOwned()` then reproduces the unified route's rule
+ * exactly — in multi-user mode a non-admin sees neither other users' sessions nor
+ * unowned host-wide history, and in single-user mode every check short-circuits
+ * true. The snapshot is written UNSCOPED, so it must never be returned unfiltered.
+ *
+ * Key exports:
+ * - setHistorySessionIndex / getHistorySessionIndex — the snapshot accessors.
+ * - buildHistorySessionIndexItems — pure merged-list → index-item projection.
+ * - setHistoryIndexRefresher / ensureHistorySessionIndexFresh — the refresh hook.
+ */
+
+/** One past-session row in the snapshot. Mirrors what the search corpus needs, nothing more. */
+export interface HistorySessionIndexItem {
+  /** Codeman session id (the search result's session id and dedupe key). */
+  sessionId: string;
+  /** Display name, may be empty for a transcript-only row. */
+  name: string;
+  /** Absolute working directory — the field issue #261 is about matching. */
+  workingDir: string;
+  /** Claude conversation UUID, when known: what a resume actually replays. */
+  claudeSessionId?: string;
+  /** Recency timestamp (lastActivityAt, else createdAt). */
+  timestamp: number;
+  /**
+   * Owning user, when the row is tied to a live or persisted session. `undefined`
+   * means host-wide transcript history, which only admins (or single-user mode)
+   * may see — the same rule `/api/sessions/unified` applies.
+   */
+  owner?: string;
+  /** True when the session is still in the live map (search harvests those directly). */
+  live: boolean;
+}
+
+/** Hard cap on snapshot size, so a host with thousands of transcripts stays bounded. */
+export const HISTORY_INDEX_MAX_ITEMS = 400;
+
+/** How long a snapshot is considered fresh before a search triggers a background refresh. */
+export const HISTORY_INDEX_TTL_MS = 60_000;
+
+interface HistorySessionIndexSnapshot {
+  items: HistorySessionIndexItem[];
+  /** Epoch ms of the last write; 0 when never populated. */
+  updatedAt: number;
+}
+
+let snapshot: HistorySessionIndexSnapshot = { items: [], updatedAt: 0 };
+let refresher: (() => Promise<void>) | null = null;
+let refreshInFlight = false;
+
+/** The merged-list shape this module projects from (a subset of `UnifiedSessionItem`). */
+export interface MergedSessionLike {
+  sessionId: string;
+  name?: string;
+  workingDir?: string;
+  claudeSessionId?: string;
+  createdAt?: number;
+  lastActivityAt?: number;
+}
+
+/**
+ * Project a merged unified list into index items. PURE — the caller supplies the
+ * owner lookup and the live-id set it already has in hand.
+ *
+ * Rows with no working directory AND no name are dropped: they can never match a
+ * query in a useful way and would only consume the cap.
+ *
+ * @param merged unified-list items, newest-first (the order the merge returns)
+ * @param ownerById owner of a session id, for rows tied to a live/persisted session
+ * @param liveIds session ids currently in the live map
+ */
+export function buildHistorySessionIndexItems(
+  merged: MergedSessionLike[],
+  ownerById: Map<string, string | undefined>,
+  liveIds: Set<string>
+): HistorySessionIndexItem[] {
+  const items: HistorySessionIndexItem[] = [];
+  for (const m of merged) {
+    if (items.length >= HISTORY_INDEX_MAX_ITEMS) break;
+    const name = m.name ?? '';
+    const workingDir = m.workingDir ?? '';
+    if (!name && !workingDir) continue;
+    items.push({
+      sessionId: m.sessionId,
+      name,
+      workingDir,
+      claudeSessionId: m.claudeSessionId,
+      timestamp: m.lastActivityAt ?? m.createdAt ?? 0,
+      owner: ownerById.get(m.sessionId),
+      live: liveIds.has(m.sessionId),
+    });
+  }
+  return items;
+}
+
+/** Replace the snapshot. Items are capped defensively even if the caller already did. */
+export function setHistorySessionIndex(items: HistorySessionIndexItem[], now = Date.now()): void {
+  snapshot = { items: items.slice(0, HISTORY_INDEX_MAX_ITEMS), updatedAt: now };
+}
+
+/**
+ * Read the snapshot. The returned array is UNSCOPED — callers must apply the
+ * per-item ownership check before exposing any of it.
+ */
+export function getHistorySessionIndex(): HistorySessionIndexSnapshot {
+  return snapshot;
+}
+
+/** True when the snapshot has never been written, or is older than the TTL. */
+export function isHistorySessionIndexStale(now = Date.now(), ttlMs = HISTORY_INDEX_TTL_MS): boolean {
+  return snapshot.updatedAt === 0 || now - snapshot.updatedAt > ttlMs;
+}
+
+/**
+ * Register the rebuild function. Called once by the session routes, which own the
+ * transcript scanner and the stores the unified list is merged from.
+ */
+export function setHistoryIndexRefresher(fn: (() => Promise<void>) | null): void {
+  refresher = fn;
+}
+
+/**
+ * Kick a background rebuild if the snapshot is stale. Returns immediately —
+ * NEVER await this from a request handler, that is the whole point: the search
+ * path answers from the current snapshot and stays free of disk I/O.
+ */
+export function ensureHistorySessionIndexFresh(now = Date.now()): void {
+  if (refreshInFlight || !refresher || !isHistorySessionIndexStale(now)) return;
+  refreshInFlight = true;
+  void refresher()
+    .catch(() => {
+      // A failed rebuild leaves the previous snapshot in place; the next search retries.
+    })
+    .finally(() => {
+      refreshInFlight = false;
+    });
+}
+
+/** Test hook: drop the snapshot and any registered refresher. */
+export function resetHistorySessionIndex(): void {
+  snapshot = { items: [], updatedAt: 0 };
+  refresher = null;
+  refreshInFlight = false;
+}