refactor(core): split search query and row hydration

[phernandez]

Why

• Begins Split core god files (search_repository_base, project_context, config, runtime protocols) #1108 with a reviewable extraction after the naming cleanup from refactor(core): rename picoschema package and drop legacy routes #1121.
• search_repository_base.py mixed backend-neutral lexical query policy and duplicated persisted-row hydration with vector search orchestration.
• The earlier Claude workflow for Split core god files (search_repository_base, project_context, config, runtime protocols) #1108 was canceled and left no branch or pull request to continue.

What Changed

• Move relaxed FTS query eligibility and tokenization into repository/search_query.py.
• Add SearchIndexRow.from_mapping() as the explicit persistence boundary used by the base, SQLite, and PostgreSQL repositories.
• Update direct imports and add regression coverage for database value normalization and single-term CJK queries.
• Preserve query SQL, ranking, fallback order, and vector orchestration behavior.

Implementation Details

• The lexical policy remains a small set of pure functions rather than a new manager or mixin hierarchy.
• Row hydration normalizes SQLite JSON strings, PostgreSQL mappings, and numeric scores in one symmetric counterpart to SearchIndexRow.to_insert().
• Backend-specific syntax, metadata filtering, and savepoint behavior remain in their existing repositories.
• Vector synchronization is intentionally left for a later Split core god files (search_repository_base, project_context, config, runtime protocols) #1108 slice.

Testing

Automated

• just fast-check — passed (existing Python 3.14 asyncio deprecation warnings only)
• BASIC_MEMORY_ENV=test uv run pytest tests/repository/test_search_relaxation.py tests/repository/test_search_index_row.py tests/repository/test_semantic_search_base.py tests/repository/test_search_repository.py tests/repository/test_postgres_search_repository_unit.py tests/services/test_search_service.py --no-cov -q — 164 passed
• just fast-test — 4,580 passed, 41 skipped, 5 failed: one live OpenAI semantic-quality case hit 429 insufficient_quota; four Codex-plugin dependency-floor cases reflected concurrent, unstaged workspace changes outside this PR

Manual

• git diff --cached --check — passed
• Verified the commit contains only the eight search repository/service/test paths listed in this PR.

Risks / Follow-ups

• The main risk is SQLite/PostgreSQL row-hydration parity; the impacted repository and service suites cover both paths.
• Continue Split core god files (search_repository_base, project_context, config, runtime protocols) #1108 with a separate vector-sync extraction rather than expanding this review.
• Concurrent Codex-plugin edits in the shared checkout were explicitly excluded from the commit.

[phernandez]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Keep it up!

Reviewed commit: 1c4bbd3ea9

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".