Add SQLite support alongside PostgreSQL for flexible deployment

[GoJam11]

Summary

This PR adds SQLite as an alternative database backend, allowing the application to run without requiring a separate PostgreSQL instance. The target database is automatically selected at startup based on the DATABASE_URL scheme and is fixed for the lifetime of the deployment to prevent data fragmentation.

Key Changes

• Dual-dialect schema structure: Split database schema into schema.pg.ts (PostgreSQL) and schema.sqlite.ts (SQLite) with identical column names, defaults, and indexes. The main schema.ts exports the appropriate dialect at runtime based on DATABASE_URL.

• Database driver detection: Added getDbDriver(), isSqliteUrl(), and getSqliteFilePath() utilities in src/db/config.ts to determine the target database from the connection string (sqlite: / file: schemes → SQLite; postgresql:// → PostgreSQL).

• SQLite client initialization: Updated src/db/client.ts to create and manage a shared bun:sqlite connection with WAL mode and busy timeout for concurrent access, mirroring the PostgreSQL connection pooling pattern.

• Migration support for both dialects: Extended src/db/migrate.ts to run SQLite migrations from drizzle/sqlite/ folder alongside PostgreSQL migrations from drizzle/. Added initial SQLite migration (drizzle/sqlite/0000_many_the_hand.sql) and metadata snapshot.

• Drizzle configuration: Updated drizzle.config.ts to select schema and output folder based on the detected driver, enabling bun run db:generate to work with either database.

• Docker Compose for SQLite: Added docker-compose.sqlite.yml for zero-dependency SQLite deployments using a named volume for persistence.

• Documentation and examples: Updated README files (English and Chinese) and .env.example to document SQLite as a deployment option. Added src/db/README.md explaining the dual-dialect architecture and migration workflow.

• Gitignore updates: Added patterns for SQLite database files (.db, .db-shm, .db-wal, .sqlite, .sqlite3).

Implementation Details

• Type mapping: PostgreSQL bigint({mode:'number'}) → SQLite integer({mode:'number'}) for 64-bit integers; serial → integer(primaryKey autoIncrement).
• The database choice is immutable at runtime; switching requires a fresh deployment to avoid split data.
• SQLite uses embedded mode with no separate connection pool; concurrency is serialized by the driver with WAL mode preventing spurious SQLITE_BUSY errors.
• All store-layer code (console-store.ts, etc.) remains unchanged; it imports from the unified schema.ts which provides the correct dialect's tables.

https://claude.ai/code/session_01U5dS4D3toAwqRG63xF97p1