FileSink: settle the pending write()'s promise in every synchronous close()/end() flush arm - #35365

The `drainedArm` sub-test assumes `close()`'s flush drains the whole remaining buffer in one shot, but that depends on the platform's AF_UNIX buffer size relative to the 300KB payload. On macOS (`net.local.stream.sendspace` defaults to 8192) `flush()` returns `Pending` — not `Done`/`Wrote` — so `run_pending_later()` is never scheduled, nobody drains `readFd` during the 500ms await, and `settled` stays `"timeout"` → the assertion fails on macOS CI. Either drop the drained-arm sub-test (the errArm

`end_from_js()`'s `Done` and `Wrote` arms have the exact same bug this PR fixes in `end()`: they call `writer.end()` and return a plain `js_number` without checking `self.pending` or scheduling `run_pending`, so a backpressured `write()`'s promise is stranded forever. The PR's own `drainedArm` test with `sink.end()` in place of `sink.close()` triggers it, and the new comment here claims the fix "mirrors … `end_from_js`" — but `end_from_js` only mirrors the `Err` arm. Per REVIEW.md's fix-the-whol