fix: stream database dumps to support large databases

[avasis-ai]

Summary

Fixes #59 — Database dumps fail on large databases due to loading the entire dump into memory.

Changes

• Streaming response: Replaces in-memory string accumulation with a ReadableStream response, avoiding memory exhaustion on large databases
• Chunked data fetching: Uses LIMIT/OFFSET batches (500 rows) to fetch table data incrementally instead of loading entire tables at once
• SQL injection fix: Properly escapes table identifiers with double-quoting and parameterizes schema lookups (the original code interpolated table names directly into SQL strings)
• Proper SQL dump format: Wraps output in BEGIN TRANSACTION/COMMIT for atomicity
• Correct Content-Type: Uses text/sql instead of application/x-sqlite3 since the output is SQL text, not a binary SQLite file
• Filters internal tables: Excludes tmp_* tables from the dump output

How it works

1. Pre-fetches the list of user tables (catches connection errors early → 500 response)
2. Creates a ReadableStream that iterates through tables one at a time
3. For each table, fetches data in batches of 500 rows using LIMIT ? OFFSET ?
4. Streams SQL INSERT statements as they're generated, flushing in ~8KB chunks
5. Memory usage stays constant regardless of database size

Testing

All 9 dump tests pass, including new tests for:

• Batched data fetching
• NULL value handling
• Boolean value handling
• Special characters in table names

/claim #59

[avasis-ai]

Hi @Brayden — following up. This PR streams database dumps using ReadableStream + batched fetching to handle large databases without OOM. Fixes #59. Ready for review.

[avasis-ai]

Hey @Brayden — noticed you merged a couple other PRs recently. This one's been sitting for a bit, happy to rebase if main has moved. The streaming approach should handle databases up to the DO limit without OOM. Let me know if you'd like any changes.