Persistent indexer storage (PostgreSQL backend)

[LiranCohen]

Summary

Replace the in-memory `IndexerStore` with a persistent storage backend (PostgreSQL) for production deployments.

Context

The current `IndexerStore` (`src/indexer/store.ts`) uses in-memory `Map`/`Set` collections with FIFO eviction (added in the hardening PR #22). This is fine for development but loses all indexed data on restart. A production indexer needs persistent storage.

Scope

• Create `IndexerStorePostgres` implementing the same public API as `IndexerStore`
• Use the existing PostgreSQL infrastructure from `docker-compose.test.yaml`
• Add database migrations for the indexer schema (repos, stars, follows, cursors)
• Make the store backend configurable via environment variable (`DWN_GIT_INDEXER_STORE=postgres`)
• Keep in-memory store as default for development

Design considerations

• The store interface is already well-defined — `putRepo`, `putStar`, `putFollow`, `getRepo`, `search`, etc.
• Full-text search (`search()` method) maps naturally to PostgreSQL `tsvector`/`tsquery`
• Trending calculation can use SQL window functions
• Connection pooling for the crawl loop + API server