fix(extractors): store state in brain DB, not JSON files (closes #63)

[rajkripal]

Problem

Extractor state files lived at <data_dir>/extractor_state/<name>.json, independent of graph.db. Wiping the DB while keeping the state dir caused silent skip-all on reingest: 0 nodes, no error.

Solution

Move extractor state into the brain DB as an extractor_state table:

CREATE TABLE extractor_state (
    extractor_name TEXT PRIMARY KEY,
    state_json     TEXT NOT NULL,
    updated_at     TEXT NOT NULL
)

State lifecycle now matches DB lifecycle. Wipe the DB, state is gone, reingest works.

Changes

• ExtractorRegistry.__init__ accepts optional db_path; creates the table on init
• _save_state / _load_state use DB when db_path is set, JSON file otherwise
• Backwards-compat shim: on first _load_state miss, checks for legacy JSON file and imports it into the DB
• cmd_ingest in cashew_cli.py passes db_path to registry
• Four new tests: DB persistence, DB lifecycle matches DB wipe, JSON fallback (no db_path), legacy shim import

Test results

38 passed in tests/test_extractors.py
285 passed overall (1 pre-existing failure in test_prepare_ingest.py unrelated to this PR)