feat: PostgreSQL/pgvector database backend

[RobertLD]

Summary

Add PostgreSQL with pgvector as an alternative database backend to SQLite, enabling multi-user concurrent access and horizontal scaling.

Motivation

SQLite is perfect for personal use but has single-writer limitations. PostgreSQL with pgvector unlocks: concurrent multi-user access, team deployments, cloud-hosted instances, and is required for any SaaS/hosted offering.

Implementation

• src/db/postgres.ts — PostgreSQL connection manager using pg (node-postgres)
• src/db/backend.ts — Abstract database interface that both SQLite and Postgres implement
• src/db/migrations-pg.ts — PostgreSQL-specific schema migrations (pgvector types, GIN indexes)
• Refactor src/db/schema.ts to be backend-agnostic
• Config: LIBSCOPE_DB_URL=postgresql://... or libscope config set db.backend postgres

Key Differences from SQLite

• Vector search via pgvector <-> (cosine distance) instead of sqlite-vec
• FTS via tsvector + ts_query instead of FTS5
• Connection pooling (pg Pool)
• Proper concurrent transactions

Migration Path

• libscope migrate --to postgres --url <pg-url> — migrate existing SQLite data to Postgres
• Auto-detect backend from connection string

Acceptance Criteria

• Database interface abstraction (SQLite + Postgres implement same interface)
• Postgres connection manager with pooling
• All schema migrations ported to Postgres/pgvector
• Vector search working with pgvector
• Full-text search working with tsvector
• Migration tool from SQLite → Postgres
• All existing tests pass against both backends
• Integration test suite for Postgres (can use Docker)