fix(searcher): guard CLI divergence before Chroma open

[igorls]

What changed

• Probe Chroma HNSW capacity before get_collection() in CLI search.
• Route diverged Chroma palaces to the existing SQLite BM25 fallback without constructing a Chroma client.
• Keep healthy Chroma and every non-Chroma backend on their normal vector path.
• Render fallback source names consistently with normal CLI results.

Why

#1967 guarded only col.query(), after collection opening. That ordering still allowed Chroma client construction and embedder-identity enforcement (including a possible collection.count()) to touch damaged native index state before the guard ran. The MCP path already probes before _get_client; the CLI path needs the same boundary.

This draft supersedes #1967 with the pre-open ordering fix while retaining its useful BM25 fallback. Original design and implementation credit: @trek-e.

Local validation

• uv run pytest tests/test_searcher.py tests/test_hnsw_capacity.py tests/test_backends.py -q — 169 passed
• Focused pre-open/healthy/non-Chroma guard tests — 3 passed
• Ruff check and format checks pass
• Real disposable-palace smoke on this Mac: created a Chroma palace with an actual drawer and SQLite FTS5 data, forced the divergence gate, made get_collection() raise if called, and confirmed CLI search returned the exact stored sentence through BM25 without opening Chroma.

[Copilot]

Pull request overview

This PR aligns the CLI search() path with the MCP path by probing for Chroma HNSW divergence before attempting to open a Chroma collection, and falls back to the existing SQLite BM25-only search when divergence is detected—avoiding native crashes while keeping output formatting consistent.

Changes:

• Add a pre-open Chroma HNSW divergence probe for CLI search and route diverged palaces to BM25-only output.
• Introduce a dedicated CLI BM25-only result printer that mirrors the normal vector-path output shape and includes an actionable repair notice.
• Add regression tests covering diverged/healthy Chroma paths and ensuring non-Chroma backends don’t invoke the Chroma-specific probe.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
mempalace/searcher.py	Adds pre-open HNSW divergence gating and a BM25-only CLI fallback printer to avoid Chroma open/query crashes.
tests/test_searcher.py	Adds regression coverage for diverged vs healthy Chroma behavior and ensures other backends aren’t fenced by the Chroma probe.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.