feat(search): surface camelCase symbols via segment vocab (#1520)

[terrabeckhub]

Summary

Addresses the largest finding in #1520: camelCase / PascalCase identifiers are unreachable by sub-word FTS queries because nodes_fts uses the default unicode61 tokenizer (no case-boundary splits).

name_segment_vocab already materializes those sub-words at index time for the prompt hook. This PR merges that vocabulary into searchNodes candidate generation so queries like checkout surface getShippingMethodIdFromCheckout without a custom FTS5 tokenizer or schema migration.

Changes

• After FTS, call supplementWithSegmentMatches() to pull symbol names from name_segment_vocab for each query term and join them into the candidate set (still subject to kind/language filters and the existing rescoring path).
• Add a regression test that indexes a camelCase helper and asserts searchNodes('checkout') finds it.

Other #1520 items (persisted bm25 rank weights, dropping id from FTS, --path CLI, GROUP BY dedupe) are left for follow-ups; search already applies column-weighted bm25(...) in SQL and supports path: query filters.

Test plan

• npx vitest run __tests__/search-segment-supplement.test.ts
• CI