Node-host timer pump can lose a delayed callback in a long-running corpus

[pmatos]

Summary

In the Zod v4.4.3 normal+jitless library corpus, jsse can exit successfully while a delayed setTimeout callback is still required to settle an awaited async refinement. Node continues and completes the test.

Reproduction

From the issue #279 branch/PR, remove the 500 to 0 portability replacement in scripts/gen-zod-entry.js, then run:

./scripts/run-library-tests.sh zod --clean

Normal mode completes the upstream async validation multiple errors 2 case. In jitless mode the corpus reports through case 1,133, schedules the case-1,134 refinement timer for 500 ms, and then jsse exits 0 without reporting case 1,134 or a TAP summary. Node reports all 2,184 cases.

The forced shared test harness implements cancellable timers using one native jsse setTimeout pump, rearming at up to 100 ms intervals. The pending native timer should keep drain_microtasks_until_idle alive until the callback resolves the awaited refinement.

Expected

Pending timer jobs remain live, their callbacks execute, and the awaiting promise continuation resumes even late in a long-running bundle.

Current corpus treatment

Issue #279 symmetrically changes only this artificial 500 ms test delay to 0 ms on Node and jsse. The timer callback and async multi-error assertion still execute, while avoiding the liveness defect. Remove that replacement when this issue is fixed.

[pmatos]

Found while re-running the library-harness battery for #415 (nan-box Phase 4
cleanup, unrelated to this issue's own cause) — reproduced against current
main (abd3b0f, i.e. including the nan-box swap, PR #439) with
./scripts/run-library-tests.sh zod, both inside a heavily-loaded 18-suite
parallel battery and again standalone on an otherwise-quiet host (load
average ~8 against 61 cores), so not a resource-contention artifact.

Current symptom is a complete zero-case wipeout in both normal and
jitless mode, not the single-late-case loss this issue's own repro describes:

$ timeout 60 ./target/release/jsse --node /tmp/jsse-libtests/zod/final-0.cjs
TAP version 13
$ echo $?
0

Exit code 0, no stderr, well under the 60s timeout — jsse's event loop
appears to conclude there is no more pending work immediately after printing
the TAP header, before running even case 1. This matches the shape
03df6fe's/#340's closing comment already flagged and deliberately deferred
as a separate, "#310-style" host-timer early exit ("two TAP headers, zero
cases, exit 0") — but where that comment implied a late/partial loss
consistent with this issue's own title, what's reproducing now is total: no
synchronous, immediately-resolvable cases print either.

Not investigating further — out of scope for #415's cleanup PR (which has
zero production codegen changes) — but flagging since it's a materially
different/worse severity than the "2,176/2,184 passing" figure previously
recorded for this suite, in case that's useful signal for whoever picks this
up next.