Add podtx search for full-text search across transcripts

[frarredondo]

Summary

After syncing multiple feeds (~1k+ episodes), finding a topic means grepping files by hand. Add a first-class search command over stored transcript text.

Motivation / evidence

Local library is already large enough that ad-hoc rg across nested feed folders is clumsy (paths, snippets, feed filters).

Proposal

podtx search "bitter lesson"
podtx search "diarization" --feed corecursive-coding-stories
podtx search "MLIR" --limit 20

Suggested behavior:

• Full-text index over transcript text (SQLite FTS5 is a natural fit next to state.db)
• Results show feed, episode title/date, short snippet, and path to .txt/.json
• Filters: --feed, date range, --limit
• Incremental indexing on sync/format, plus optional podtx search --reindex

Out of scope

Semantic/vector search, web UI, summarizing search results.

Acceptance criteria

• Search returns relevant hits from already-transcribed episodes
• Works offline against local data
• Documented in README with examples