fix(search): match snake_case identifiers in BM25 queries

[sonwr]

This fixes BM25 search for snake_case identifiers such as atomic_write_json.

What changed:

• replace non-alphanumeric characters with spaces instead of removing them during FTS5 term sanitization
• align the CLI-side sanitization logic with the store implementation
• add an end-to-end CLI test covering snake_case identifier search
• update structured search tests for the new tokenization behavior

Why:
SQLite FTS5 with the unicode61 tokenizer splits snake_case into separate tokens at index time. The previous query sanitization removed underscores, turning atomic_write_json into atomicwritejson, which could not match the indexed tokens.

How I tested it:

• reproduced locally before the change:
  • search "atomic_write_json" returned []
  • search "atomic write json" returned the expected document
• verified locally after the change:
  • search "atomic_write_json" returns the expected document
  • search "parse_http_response" returns the expected document
• ran targeted tests:
  • test/structured-search.test.ts
  • test/cli.test.ts

Fixes #305

[sonwr]

Thanks. This change is small and still looks directionally fine, but the branch has gone stale and is now conflicting with the base branch. Rather than keep a conflicting PR open without active reviewer traction, I am closing this out here. If it still makes sense against current main later, I would prefer to reopen it as a fresh rebase or smaller follow-up.