refactor(core): compile search filters over an explicit ProjectScope

[phernandez]

Why

Part 1 of the #1558 stack. Single-project search, the database-scoped search route (the contract from #1559 and cloud #2080), and MCP search_all_projects should be one read pipeline parameterized by project scope, so one project ranks identically through every route. This PR makes the scope an explicit value and moves FTS compilation out of the project-bound repositories so a shared reader can call it next.

No query behavior changes. Project repositories call the same compilers with a scope of one.

What changed

• search_scope.ProjectScope: sorted, unique, positive project IDs, built once at the boundary with of() or single(). predicate(column, params) renders column IN (:scope_0, ...) and binds each ID once per statement; an empty scope renders 1 = 0. Absence is not representable.
• search_filters: FilterDialect holds the two spellings that differ per backend (note-type JSON accessor, after_date comparison). CompiledFilter replaces the 5-tuple the compilers returned. shared_filter_conditions compiles the filters both backends previously carried as copies: scope, permalink, directory subtree, candidate keys, item type, category, note type, after_date, valid time.
• sqlite_search_query / postgres_search_query: term preparation, relaxed renderers, syntax-error classifiers, and compile_fts_filter as module functions. The SQLite compiler is pure; the repository resolves the entity table's columns and passes them in.
• note_type_filters / temporal_filters: take scope= and match search rows on their full (project_id, ...) identity.
• SearchRepositoryBase: gains self.scope; drops the abstract _prepare_search_term hook nothing in the base called.
• Tests: call the module functions directly through a small fts_query(search_repository) helper; Postgres tests patch the syntax-error classifier on the repository module where it is bound. New test_search_scope.py.

Not in this PR

• PR 2: extract vector retrieval, fusion, relaxation retry, and chunk hydration into a scope-parameterized SearchReader; project search() delegates.
• PR 3: QUERY /v2/search/ and the scoped service method, porting the feat(core): add explicit multi-project API search #1559 tests.
• PR 4: vector order per request shape and ANN pre-filter, with a tenant benchmark.
• PR 5: back MCP search_all_projects with the scoped reader.

This supersedes the extraction half of #1559. That PR should not merge as-is.

Testing

Check	Result
just fast-check	passed
uv run pytest tests/repository tests/services/test_search*.py tests/api/v2/test_search*.py tests/api/v2/test_cjk_search_guidance.py --no-cov (SQLite)	1277 passed, 38 skipped
BASIC_MEMORY_TEST_POSTGRES=1 uv run pytest over the search repository, relaxation, temporal, metadata-filter, and search router suites (testcontainers)	300 passed, 15 skipped
just doctor	passed

🤖 Generated with Claude Code

https://claude.ai/code/session_019YW9ysxugGGBCNEGzsxtFV

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.