feat: stream-based database export for large databases

[chaudl113]

Fixes #59

Summary

Replace in-memory database dump with streaming approach to support large databases (10GB+) without hitting the 30s Durable Objects timeout.

Changes

• TransformStream for chunked response delivery — data streams to client as it's generated
• Paginated queries (LIMIT/OFFSET, 500 rows per batch) — no full table load into memory
• Breathing intervals (50ms between batches) — prevents DO lockup, allows other requests to process
• Proper value escaping — handles strings, numbers, NULLs, BigInts, and binary blobs (hex encoding)
• Backward compatible — same endpoint (/export/dump), same headers, same response format

How it works

1. Table list fetched synchronously (fail fast if DB is broken)
2. Background IIFE streams data via TransformStream
3. Each table: schema first, then data in 500-row batches
4. 50ms breathing pause between batches yields to DO event loop
5. Client receives chunked Transfer-Encoding response

Testing

• All 5 existing test cases pass (adapted for streaming)
• Verified: multi-table, empty DB, no-data, quote escaping, error handling

/claim #59