Stream large database exports

[sureshchouksey8]

/claim #59

Summary

Fixes #59 with a shared streaming export path for the SQL dump, CSV, and JSON export endpoints so large tables are exported in bounded pages instead of being materialized into one in-memory payload.

What changed

• Added shared export streaming helpers for:
  • safe SQL identifier quoting
  • SQL value serialization, including NULL, booleans, finite numbers, bigint, and binary values
  • CSV value escaping
  • sanitized download filenames
  • bounded table-row iteration with runtime yielding
• Updated /export/dump to stream schema and INSERT statements progressively.
• Updated /export/csv/:tableName and /export/json/:tableName to stream rows progressively instead of loading the whole table before creating the response.
• Uses deterministic pagination order: primary-key columns when available, falling back to rowid for tables without a primary key.
• Keeps table existence checks parameterized and excludes SQLite internal tables from full database dumps.
• Added focused coverage for the helpers plus SQL/CSV/JSON export routes.

Why this should help large databases

The previous implementation loaded all exported rows into memory before sending a response. This branch sends a ReadableStream response and reads rows in bounded pages, which keeps memory bounded and starts returning bytes immediately for large exports.

Verification

• npx -y pnpm@9.15.4 install --frozen-lockfile --ignore-scripts
• npx -y pnpm@9.15.4 exec vitest --run src/export ✅ 5 files / 34 tests
• git diff --check ✅

I also ran npx -y pnpm@9.15.4 exec tsc --noEmit --pretty false; it still reports existing unrelated project type errors in CDC/cache/DO/operation tests, with no errors from the export files changed here.