test: force mocha to exit so a leaked handle cannot hang CI

[petersalomonsen]

What happened

The Publish job on a91915a (run 32522680732) sat for 6 hours and was killed by GitHub's hard limit.

The trigger was a transient network failure, not a code problem. test/nodefs.spec.js clones wasm-git from github.com over the real internet, and that transfer died at 68%:

net 68% ( 247 kb,   476/  699)  /  idx  40% (  285/  699)
Bad news:
 could not read from remote repository
ERROR 12: could not read from remote repository

The ErrnoError errno 44 mocha reported is just the follow-on ENOENT from FS.chdir() into a clone directory that was never created. The CI workflow on the same commit passed in 6m27s, running the same node suite — so this was a flake, not a regression.

The damage came afterwards: mocha printed 13 passing, 1 failing and then never exited. The aborted clone leaves a live handle in the emscripten module, and with no .mocharc and no --exit the event loop never drains. The job held a runner until the 6-hour cap, ending in Terminate orphan process: pid (12095) (npm run test).

The change

--exit on the two mocha scripts, so the process terminates once the run is over regardless of what handles third-party code left behind.

Verification

A probe that leaks a listening socket and throws, run under a 10-second cap:

	exit code
mocha	124 — killed by the cap, still hanging
mocha --exit	1 — exits immediately, failure preserved

Both node suites still pass: npm test (14) and npm run test-opfs-detect (8).

Scope

This treats the symptom. The handle leak in the aborted clone, and the fact that nodefs.spec.js depends on reaching github.com at all, are both still there — worth a retry wrapper or a local fixture if that test keeps flaking. What this buys is that a hang becomes an immediate visible failure instead of an occupied runner.

Complements #122, which caps every job at timeout-minutes: 30.

🤖 Generated with Claude Code