feat(transcripts): persist transcript index entries in SQLite

[EloyEMC]

Problem / requested outcome

The roadmap places local transcript indexing before embeddings and semantic search. Issue #19 defines deterministic in-memory TranscriptIndexEntry values, but the repository has no persistence boundary. Persist those entries using the existing SQLite/libSQL and Drizzle conventions, without coupling storage to embeddings or search.

Scope

• Add an idempotent transcript-index table through existing database initialization conventions.
• Persist video ID, chunk index, exact text, deterministic identity, and input order.
• Preserve duplicate entries; identity alone must not be the sole row key.
• Use conservative local-install scope because the current entry contract carries no user or channel ownership.
• Define replace-per-video write semantics as an atomic operation: the replacement must not leave a partial index on failure.
• Provide a narrow repository/adapter for replacing entries for a video, retrieving entries in deterministic order, and deleting entries for a video.
• Keep the adapter compatible with the exact TranscriptIndexEntry contract from issue feat(transcripts): add storage-neutral index builder #19 commit 77cffa8.

Acceptance criteria

• Schema and initialization follow existing Drizzle/libSQL conventions and are idempotent.
• Replace, list, and delete operations are typed and covered by focused tests.
• Replacing a video index removes stale entries and writes the new ordered entries atomically.
• Duplicate input entries remain distinct and retain their order.
• Retrieval is deterministic and preserves stored text, video ID, chunk index, identity, and order.
• Empty replacement has explicit tested behavior.
• No user/channel ownership is invented or persisted.
• Existing transcript, chunking, index-builder, database, and full tests remain passing.

Non-goals

• Embedding provider integration or vector columns.
• Vector or semantic search, lexical ranking, or retrieval APIs.
• Transcript fetching, fallback, normalization, or chunking changes.
• API, CLI, MCP, UI, quota, analytics, playlist, or background-job integration.
• Backfill, migrations for existing transcript data, or provider/model configuration.
• Changing the public TranscriptResult or TranscriptIndexEntry contracts.

Planned validation

• Focused storage tests using existing database test conventions.
• npm test
• npx tsc --noEmit
• npm run lint
• git diff --check
• npm run build, subject to the known external node_modules symlink/Turbopack issue.

Dependency: issue #19, commit 77cffa8.
Roadmap: Phase 4 local transcript-index persistence before embeddings and semantic search.

[EloyEMC]

Implemented and committed as f7b4425 (feat(transcripts): persist index entries in SQLite). Added idempotent Drizzle/libSQL schema and a typed storage adapter with atomic replace-per-video, deterministic listing, delete-by-video, empty replacement, and duplicate preservation. Validation: 14 focused storage/index/chunker tests passed, 191/191 full tests passed, TypeScript passed, ESLint passed, and git diff check passed. npm run build remains blocked by the external node_modules symlink/Turbopack error: Symlink project node_modules is invalid, it points out of the filesystem root. No push or PR was created.