refactor(core): extract semantic chunk planning

[phernandez]

Why

• Continue Split core god files (search_repository_base, project_context, config, runtime protocols) #1108 with the next bounded SearchRepositoryBase extraction after refactor(core): split search query and row hydration #1122 split query and row-hydration concerns.
• Semantic source composition, Markdown-aware chunking, deterministic chunk records, and entity fingerprints are pure planning work. Keeping them inside the async repository base mixed deterministic transformations with database and embedding orchestration.

What Changed

• Added repository/semantic_chunking.py as the typed, side-effect-free home for semantic chunk planning.
• Kept thin compatibility delegates on SearchRepositoryBase while removing the planning implementation from the repository class.
• Preserved project-aware duplicate chunk logging in the repository orchestration layer.
• Updated the Postgres vector scheduling hook to use the shared VectorChunkRecord type.
• Moved pure chunk-planning tests to the new module boundary and expanded edge-case and fingerprint coverage.
• No user-visible semantic search behavior changes are intended.

Implementation Details

• SemanticSourceRow defines the narrow search-row input contract.
• VectorChunkRecord names the deterministic writer input, and VectorChunkBuildResult returns records with duplicate-key diagnostics.
• Source composition, chunk splitting, record hashing, and entity fingerprinting are ordinary module functions with no repository or database dependencies.
• SearchRepositoryBase still owns backend coordination, embedding calls, state inspection, logging, and persistence hooks.

Testing

Automated

• just fast-check: passed (Ruff formatting/lint and ty type checking).
• BASIC_MEMORY_ENV=test uv run pytest tests/repository/test_semantic_chunking.py tests/repository/test_semantic_search_base.py tests/repository/test_sqlite_vector_search_repository.py tests/repository/test_postgres_search_repository.py test-int/semantic/test_search_diagnostics.py --no-cov -q: 51 passed, 23 skipped after rebasing onto current main.
• just fast-test: 4,594 passed, 41 skipped.
• BASIC_MEMORY_ENV=test uv run pytest tests/repository/test_semantic_chunking.py tests/repository/test_semantic_search_base.py --cov=basic_memory.repository.semantic_chunking --cov-report=term-missing:skip-covered --cov-fail-under=0 -q: 33 passed; semantic_chunking.py reached 141/141 statements (100%).
• git diff --check: passed.

Manual

• just doctor: passed the local configuration, database, file/index/search, and FastEmbed vector-sync health checks.

Risks / Follow-ups

• This is a behavior-preserving refactor; the primary risk is accidental drift in chunk boundaries or hashes. Existing backend tests, the full suite, and direct 100% module coverage guard those contracts.
• This PR is part of Split core god files (search_repository_base, project_context, config, runtime protocols) #1108 and intentionally does not close it. Vector-sync orchestration and the other tracker-listed core files remain follow-up slices.