event_loop: don't debug-panic on late main-VM enqueue_task_concurrent during process.exit - #36020

The test's race trigger relies on `__cxa_atexit` running `do_close_and_sleep` after `destroy()`, but on non-ASAN Linux builds `bun_core::Global::exit()` calls `quick_exit()` (which skips `__cxa_atexit` handlers) instead of `libc_exit()` — so on `bun bd --asan=off` the write is never unblocked and the test passes vacuously even with the fix reverted. This is fine in practice (default `bun bd` on Linux and the linux-x64-asan CI lane both enable ASAN, so the test is live where it matters), but cons

The child fixture's `setTimeout(..., 50)` before `process.exit(0)` needs a comment naming why no observable signal exists (per REVIEW.md's ≥50ms rule) — but one *does* exist: since `rd` is opened `O_NONBLOCK`, a bounded poll of `fs.readSync(rd, Buffer.alloc(1))` (catching EAGAIN) until it returns >0 proves the work-pool thread has entered `write()` and is blocked. Consider replacing the fixed 50ms with that poll, or adding the comment; the 300ms atexit `usleep` makes actual flakiness unlikely, s