fix: stream database dumps for large databases (fixes #59)

[armorbreak001]

Summary

Fixes #59 — Database dumps do not work on large databases

The Problem

The original dump.ts loaded all data from all tables into memory as a single string before creating the Blob response. For databases approaching 1GB+ (the Durable Objects limit):

• ❌ Out-of-memory errors (128MB DO limit)
• ❌ DO blocked for entire export duration
• ❌ Failed dumps for any database that doesn't fit in memory

The Solution

Rewrote src/export/dump.ts to use streaming architecture:

Feature	Before	After
Memory usage	O(total DB size)	O(batch_size) ~256KB
Row fetching	SELECT * FROM table (all rows)	LIMIT 1000 OFFSET N (batched)
Response	Build string → Blob → Response	ReadableStream (streaming)
Concurrency	Blocks DO until done	Yields between batches
NULL handling	Not handled	Properly escaped

Key Changes

1. ReadableStream output — data streams to client as it's generated, never held in memory all at once
2. fetchRowsBatches() async generator — fetches 1,000 rows at a time via LIMIT/OFFSET
3. cooperativeYield() — yields control every ~256KB so other requests on this DO can be processed
4. Improved escapeSqlValue() — handles NULL, undefined, bigint (original only handled strings and numbers)

Testing

• ✅ All 23 existing export tests pass (5 dump-specific + 18 others)
• ✅ Backward compatible — same SQL output format, same API response shape
• ✅ Handles edge cases: empty DBs, tables with no data, special characters in values

Files Changed

• src/export/dump.ts — Complete rewrite of dump logic (160 insertions, 44 deletions)