fix(export): stream database dump in batches for large DBs #59

[FishEnjoyer2025]

Fixes #59 — database dumps failing on large databases.

Problem

dumpDatabaseRoute assembled the entire dump in a single in-memory string and read each
table with one unbounded SELECT *. On large databases this exhausts memory and/or
exceeds the 30s request window.

Fix

• Stream the response via a ReadableStream, so bytes flow to the client as they're
  produced — the connection stays active instead of buffering one giant body, which keeps
  the request from hitting the 30s wall.
• Read each table in bounded batches (LIMIT/OFFSET, default 1000 rows), so memory
  usage stays roughly constant regardless of table size.
• The initial table-list query stays up front so an early failure still returns a clean
  500 (rather than a half-streamed 200). batchSize is an optional parameter for tuning.

Tests

All existing dump.test.ts cases still pass, plus a new one verifying that a table larger
than the batch size is streamed across multiple OFFSET queries. 6/6 passing.

Notes

This solves the common case (constant memory + no 30s timeout) with no intermediate
storage. For truly massive (10GB+) dumps that could exceed a Durable Object's execution
budget even while streaming, a follow-up could spool to R2 and return a download URL —
happy to do that as a separate PR if you'd prefer that direction.

/claim #59