- Remove dead `index_file()` function (replaced by `_parse_file_for_indexing` + `_store_parsed_file`)
- Add `symbol_id` to `hybrid_search` results so `find_definition` doesn't need ambiguous fallback lookup
- Cache file contents in `find_references()` to avoid re-reading the same file per reference
- Add LRU cache (128 entries) on `embed_text()` to skip redundant model calls for repeated queries
- Unify duplicate `SKIP_DIRS` into a single constant in `parser.py`, imported by `doc_parser.py`
- Add stale file cleanup pass in both `index_directory()` and `index_doc_directory()` to remove ghost entries for deleted files

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

… var

Add CODE_MEMORY_BATCH_SIZE environment variable (default 64) to centralize
batch size configuration. Remove hardcoded batch_size=64 from callers in
parser.py and doc_parser.py. Document device auto-detection priority
(CUDA > MPS > CPU) and CODE_MEMORY_DEVICE env var in index_codebase tool.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>