fix(searcher): scope _expand_with_neighbors by parent_drawer_id (#1580)

[ggettert]

Fixes #1580.

Summary

Implements Option B from the issue: scope _expand_with_neighbors
by parent_drawer_id (in addition to source_file) when the matched
chunk carries one.

Why this is bigger than the one-line guard relaxation

Pre-fix, the helper bailed early when source_file was empty:

if not src or not isinstance(chunk_idx, int):
    return {"text": matched_doc, "drawer_index": chunk_idx, "total_drawers": None}

That guard masked the bug on the MCP-chunked path (where every chunk
has source_file="") by silently disabling neighbor expansion for it.
So the user-visible symptom on develop today is "neighbor expansion
quietly doesn't work for tool_add_drawer oversized content"; the
real leak only surfaces if two drawers ever share a non-empty
source_file.

This PR:

1. Relaxes the early-return: bail only when both source_file and
   parent_drawer_id are missing.
2. Adds parent_drawer_id to the $and filter (neighbor query and
   the total_drawers count query) when it's present on the matched
   chunk.
3. Net effect: neighbor expansion is restored for the MCP-chunked path
   and hardened so cross-drawer leakage is impossible even if
   source_file collides.

Live verification

Beyond unit tests, I built three harnesses:

1. Adversarial collision (two drawers, shared non-empty
source_file, distinct parent_drawer_id):

	DEVELOP	BRANCH
Expansion contains ALPHA-MARKER	✅	✅
Expansion contains BETA-MARKER	❌ leak	✅ scoped out
total_drawers reported	6 (combined)	3 (correct)

2. Natural MCP path (oversized tool_add_drawer, no
source_file — Igor's exact repro):

	DEVELOP	BRANCH
Neighbor expansion runs	❌ short-circuits	✅ runs, scoped
total_drawers reported	None	2 (correct)

3. Regression fingerprint (diary chunks + small drawers + closet
hits + MCP chunked drawers, 3 query patterns):

diff develop.json branch.json — zero diff. Existing file-backed
and closet-boosted paths are byte-identical pre/post.

Test coverage

tests/test_closets.py::TestDrawerGrepExpansion:

• test_expand_scopes_by_parent_drawer_id_when_present — pins the
  searcher: _expand_with_neighbors stitches unrelated chunks across MCP drawers sharing empty source_file #1580 repro (two drawers, shared empty source_file, scoped result).
• test_expand_without_parent_drawer_id_keeps_source_file_scope —
  pins the legacy diary path keeps source_file-only scope.

Full suite: 259/259 in tests/test_closets.py + tests/test_mcp_server.py.
Ruff clean.

Out of scope

• Whether source_file should default to a per-drawer synthetic key
  on the MCP path (the issue calls this out explicitly).
• Whether to invalidate / repair palaces that already accumulated
  leak-prone metadata (no migration needed; the fix is purely at the
  query layer).

[gemini-code-assist]

Code Review

This pull request addresses issue #1580 by scoping neighbor expansion in _expand_with_neighbors using parent_drawer_id when source_file is empty or missing. This prevents chunks from unrelated drawers with empty source files from leaking into the neighbor query. The PR also updates the total drawers lookup to use the same scoped filter and adds corresponding unit tests to verify the new scoping behavior and ensure no regression on the legacy source-file-only path. There are no review comments, and I have no feedback to provide.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[ggettert]

Closing — I missed PR #1628 when scoping this work. #1628 also closes #1580 and implements the same parent_drawer_id scoping fix, covers both _expand_with_neighbors and the sibling search_memories enrichment loop via a shared _scoped_source_filter helper, and already has test coverage for both paths plus the empty-string edge case. Apologies for the duplicate noise.