test(export): guard against memory exhaustion on large CSV exports - #1928

Closes #1794

Summary

Investigated src/routes/export.ts (the transaction bulk-export route) before writing any code, per this repo's own habit of checking whether described bugs still exist. It already uses pg-query-stream + Node's stream/promises pipeline() to stream rows from Postgres straight through a Transform into the HTTP response — no SELECT * result set is ever buffered fully in memory, in any of the CSV, JSON, or PDF export paths. There is no live memory-exhaustion bug in the current export route.

What this PR adds

A regression test that can actually catch a future reintroduction of buffering:

• Streams 200,000 synthetic rows from a lazily-generated source (a generator function wrapped in Readable.from — deliberately never a real in-memory array of 200k rows, since building one would defeat the whole point of the test).
• Asserts the HTTP response arrives as many small chunks, not one/a few large ones (maxChunkBytes < totalBytes / 5, dataEventCount > 10). A buffered implementation (collecting every row into an array or one big string before writing anything) would produce a single large data event close to the full response size — this assertion would fail against that regression and passes against the current, correctly-streamed implementation.

Why not assert on process.memoryUsage() heap deltas directly

I tried it first. GC timing under Jest/V8 makes heap measurements too noisy to assert on reliably — the same passing implementation intermittently failed a heap-growth bound by 2-3x from run to run with zero code changes. The chunk-shape assertion above is the deterministic, reliable signal for "streamed vs. buffered" and is what the test actually relies on.

Disclosed, not fixed here (pre-existing, unrelated)

tests/routes/export.test.ts — a second, older test file covering the same route with a different X-API-Key/ADMIN_API_KEY auth model — hangs indefinitely on every test in it. Confirmed present on a clean upstream/main checkout with zero code changes, reproduced both with and without a real local Postgres instance configured (ruled out "just needs a database" as the cause). Given the two files clearly test two different, non-overlapping generations of this route's auth model, this looks like a stale file left behind after a refactor rather than something in active use — flagging it rather than guessing at a fix, since I couldn't pin down the actual hang cause within a reasonable amount of investigation time.

Verification

• npx jest src/routes/__tests__/export.test.ts — 7/7 passing (6 pre-existing + 1 new), run multiple times for stability.
• npx eslint on the touched file — clean.

Testing requirements checklist (from the issue)

• Implement/resolve changes associated with the memory-exhaustion concern — confirmed already resolved by the existing streaming architecture; added the missing regression coverage.
• Ensure all existing and new unit tests compile and run successfully — src/routes/__tests__/export.test.ts (7/7). Note: tests/routes/export.test.ts is a separate, pre-existing broken file (see disclosure above), unaffected by and unrelated to this change.
• Code passes lint.