fix: final checkup review follow-ups

- i18n: the spreadsheet notice's feature words (charts, drawings, macros,
  pivot tables, external links) were bare, case-insensitive zh-CN keys, so
  the page translator also renamed a charts/ or macros/ folder in the Files
  panel and a case of that name in the case picker. They are now scoped
  'Spreadsheet feature: <word>' keys; warningText() falls back to the plain
  word when the scoped key has no translation (English, no i18n). Removing
  the bare keys closes this branch's regression. The older 'models' key has
  the same class of problem; it is left alone here, since adding
  .case-combobox-option-label to USER_TEXT_SELECTOR would also untranslate
  the picker's two action rows and still miss the title attribute.
- Spreadsheet notice bar: marked data-i18n-skip. It is written already
  translated, item by item, and ends with a number format's code, which the
  observer's t() over the whole line rewrote ({name}, "Codeman").
- Connection tile (Header Stats Style Tiles): applyLocalization() now repaints
  the indicator, so a switch back to English no longer leaves the Chinese
  value word in its data-i18n-skip span until the next keystroke or ACK.
- Tiles default: loadAppSettingsFromStorage() no longer caches the
  posture-dependent showTileGridButton default, so the init merge never
  persists it; a 2-in-1 first opened as a tablet gets the button once docked.
  Every reader still resolves the absent key through a fresh
  getDefaultSettings(), so phones stay OFF and desktops ON.
- Tile grid over a split: closeSplitPane() skips Pane A's closing resize only
  when Pane A becomes a tile. With mergeSplit false (Open group as tiles, a
  stored grid) a Pane A left out of the set gets its full width back before
  the main terminal parks, instead of keeping the split's half width.
- By-case tab strip on tablets and phones: with the boxes dissolved, the
  -<case> part of a generated name shows again, so w1-alpha and w1-beta no
  longer both read "w1".

Each new assertion fails against the previous source.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

test/tile-grid-split-coexistence.test.ts

@@ -83,6 +83,7 @@ describe('opening the grid over an open split', () => {
     app.activeSessionId = 's-a';
     localStore.set('codeman:tile-count', '2');
     const paneB = openSplit(app);
+    app.sendResize.mockClear();
 
     app.toggleTileGrid();
 
@@ -91,6 +92,8 @@ describe('opening the grid over an open split', () => {
     // Filled from tab order (s-other first), not seeded with the split's two.
     expect(app._tileGrid.ids).toEqual(['s-c', 's-other']);
     expect(app.activeSessionId).toBe('s-c');
+    // Pane A (s-a) is not a tile: the split's closing resize gave it its full width back.
+    expect(app.sendResize.mock.calls).toEqual([['s-a', { force: true }]]);
   });
 
   it('an explicit open over a split keeps both split sessions first', () => {
@@ -126,6 +129,12 @@ describe('opening the grid over an open split', () => {
     expect(app._splitPane).toBeNull();
     expect(app._tileGrid.ids).toEqual(['s-c', 's-d']);
     expect(app.activeSessionId).toBe('s-c');
+    // Pane A (s-a) is not in the group, so no tile will ever size its PTY: the
+    // split's closing resize gives it its full width back before the main
+    // terminal parks, or it stays at the split's half width for as long as
+    // the grid is open. The split's opening resize, then the closing one.
+    expect(app.sendResize).toHaveBeenCalledTimes(2);
+    expect(app.sendResize).toHaveBeenLastCalledWith('s-a', { force: true });
   });
 
   it('Ctrl/Cmd+click over a split opens the remembered count in total, never one more', () => {
@@ -139,6 +148,8 @@ describe('opening the grid over an open split', () => {
     // The split's Pane A seeds the set (tileGridOpenSet), the clicked one joins: 2, not 3.
     expect(app._tileGrid.ids).toEqual(['s-a', 's-c']);
     expect(app.activeSessionId).toBe('s-c');
+    // Pane A is a tile, which sizes its PTY: no closing resize for it.
+    expect(app.sendResize).toHaveBeenCalledTimes(1); // the split's own opening resize only
   });
 
   it('Ctrl/Cmd+click over a split with a remembered grid: the remembered grid wins, not the split', () => {