feat(i18n): the shortcut rows speak Chinese

[lymanzhao]

The Shortcuts settings page was half translated: its own chrome (the subtitle, Reset all, the conflict note) had locale entries from #330, and every row under it did not. Those row labels and hints were English strings in src/lib/shortcuts.ts, so a Chinese UI drew a Chinese page whose whole list was English. The ⌘/ cheat sheet draws the same rows and was in the same state.

This moves that copy into the locale files, under settings:shortcuts.*, and adds src/lib/shortcutCopy.ts as the id to key map plus the accessors both surfaces read. One copy is the only way the settings list and the sheet cannot disagree about what a command is called, and a new ShortcutId is a compile error until it has a key there.

Shortcut ids are untouched: they are what prefs stores the binding map under, so nothing migrates. English is unchanged word for word, which keeps the specs that assert on it valid.

Pieces:

• lib/shortcutCopy.ts: a Record<ShortcutId, string> key table, the {{cmd}} style chord variables the hints interpolate, and the accessors (shortcutLabel, shortcutHint, shortcutGroupLabel, fixedShortcut*, doubleShiftModeLabel, ctrlTabModeLabel).
• lib/shortcuts.ts: the registry drops label / hint / fixedReason and keeps structure (ids, groups, bindings, glyphs). DOUBLE_SHIFT_MODES and CTRL_TAB_MODES become id lists.
• locales/{en,zh-CN}/settings.ts: copy for 43 rows, the 2 fixed gestures, the mode selects and the group headings.
• lib/shortcutCopy.test.ts: keys built from a table are invisible to usedKeys.test.ts (literals only) and to parity.test.ts (en against zh-CN only, so a row missing from BOTH passes it), which is exactly the failure that would ship here: a row drawing its own key on screen.
• e2e/specs/i18n.e2e.ts: a case that opens the page in Chinese and reads the row labels and the mode select's options back, with an English control beside it.

Two strings stay English on purpose:

• The Code navigation group heading. That name is the feature's own (lib/lsp/featureName.ts), it follows the type-checking switch, and four other surfaces call it that. A second name for it here is the drift that file exists to prevent, so localizing the feature name itself is its own change.
• "Super" in the Linux-only hint, as before.

Suites: npm test 3104 passed, npm run typecheck:e2e clean, and e2e/specs/i18n.e2e.ts green at 7 cases including the new one. The full make e2e run reported 29 of 31 spec files passing with two failing; I could not read which two (the log was tailed) and could not re-run them afterwards, because a dev build had replaced the e2e binary. The shortcut-related specs should be read as verified, the rest of that run as unverified here.

Screenshots: the e2e case writes two, .e2e/artifacts/settings-shortcuts-en.png (the English page, unchanged) and settings-shortcuts-zh.png (the Chinese one, rows and all). The upload API refuses from this setup, so they still need a drag and drop into this description.

Manually tested by hand: the Shortcuts page was opened in Chinese and reads as it should. The screenshots come from the e2e run.

Co-Authored-By: Claude Code noreply@anthropic.com

[lymanzhao]

[simion]

Verified the 44-id def/hint mapping is complete and parity-tested in both languages, and confirmed no shortcut id changed so there's no keybinding migration risk. Thanks for the contribution!