feat(plugins): add replicate plugin to pull external data into the internal DB (#72)

[trongtruong110-ux]

Problem

#72 asks for a way to pull data from an external data source (e.g. a Postgres on Supabase) into the internal Durable Object SQLite, so a StarbaseDB instance can act as a close-to-edge read replica.

Changes

Adds a new ReplicatePlugin (plugins/replicate/) — a pull-based replicator built to the requirements in the issue:

• Per-table config — choose which tables to replicate, which columns to pull, and a cursorColumn for incremental polling. With no tables configured it discovers and full-syncs every base table from the external source.
• Incremental, append-only polling — tables with a cursorColumn only pull rows newer than the last replicated value, tracked per-table in a tmp_replicate_cursors meta table.
• Keyset pagination (O(n)) — incremental batches use WHERE cursor > ? ORDER BY cursor ASC LIMIT n, with the cursor advancing every batch, so large tables are not re-scanned. LIMIT/OFFSET re-walks every skipped row (O(n²)). Tables without a cursor column fall back to LIMIT/OFFSET full re-sync.
• Bounded memory — rows are pulled one configurable batch at a time (batchSize, default 1000).
• Schema mirroring — destination tables are created in the internal DB with column types inferred across the batch (so a NULL in the first row doesn't mistype a column).
• Routes — POST /replicate/run triggers a cycle; GET /replicate/status reports per-table cursor + last-run info. Both admin-only.
• Interval pulling — runReplication() is public so it can be driven on a schedule from a CronPlugin callback.

Registered in dist/plugins.ts alongside the existing plugins.

Tests

plugins/replicate/index.test.ts covers incremental keyset pagination (asserting the cursor query advances and OFFSET is not used), full-resync OFFSET paging, value coercion (boolean / date / null), and the dataSource guards.

/claim #72