Three follow-ups from the post-merge review of #676 (Timeline →
Source tab navigation). All three address user-visible edge cases
the original PR's tests didn't cover; mechanically narrow.

1. **100-chunk limit miss → bump to 500 + miss toast.** When
   ``_renderMemorySourceTree`` consumes ``pendingActivatePath``
   alongside a ``pendingActivateChunkId``, the original PR called
   ``browseSource(path)`` with the default ``limit=100``. A target
   chunk past position 100 silently no-op'd the highlight — the
   user clicked "Open in source" and saw the source open without
   any flash, which reads as a regression from #676's old
   showDetailFromChunk path that always rendered a detail panel.
   ``_renderMemorySourceTree`` now passes ``limit=500`` only when
   the caller asked for a specific chunk. The Home/recent and
   command-palette callers (no chunkId) keep the default 100 fetch
   so they don't pay the larger payload for a source-level jump.
   ``browseSource``'s pending-chunk consumer now shows
   ``toast.chunk_target_missing`` if the card still isn't found
   after the 500-row fetch — a 500+ chunk source is rare but
   reachable, and the toast makes the no-flash visible instead of
   reading as broken.

2. **Stale ``pendingActivateChunkId`` across manual source pivots.**
   If the user clicked "Open in source" in Timeline and then —
   before the Sources render finished — manually clicked a
   different source-item in the tree, the prior chunkId stayed in
   STATE and would silently consume on the next ``browseSource``
   render of any source. Functional impact was nil (chunk ids are
   UUIDs, so no wrong card matched) but the next legitimate
   Timeline jump would land on a state-machine-warm path. Manual
   ``.source-item`` click handler now clears
   ``STATE.pendingActivateChunkId`` as the first line, matching the
   "explicit user pivot resets the navigation hint" pattern already
   in place for the namespace chip.

3. **Hard-coded "Copy" button label in Timeline chunks-view.** The
   sibling "Open in source" button moved to ``data-i18n`` in #676
   but the adjacent ``.tl-btn-copy`` next to it stayed hard-coded
   English. Same row, same i18n surface, same locale parity rules
   from #29 and #587 — promoting it to ``timeline.copy`` (en
   "Copy" / ko "복사") closes the gap.

Cache-bust: ``app.js?v=88→89``, ``timeline.js?v=2→3`` per
``feedback_static_asset_cache_bust.md``. ``style.css`` unchanged.

Verified against ``mm web`` with Playwright MCP:
  - Timeline → Open in source fetches ``/api/chunks?source=…&limit=500``
    when ``pendingActivateChunkId`` is set (i.e. always for the
    chunks-view "Open in source" + files-view per-chunk paths).
  - Copy button renders "복사" on KO and ``data-i18n="timeline.copy"``
    so the langchange hook keeps it in sync on toggle.
  - Manual source-item click clears the stale chunkId before
    ``browseSource`` runs, verified by reading the click handler's
    first statement.

i18n parity (``test_i18n.py`` 12 passed) and ``ruff`` clean.

Co-Authored-By: Claude <[email protected]>