fix(event-loop): release job wrapper ref after scheduling

[steinnes]

What

Add the missing Py_DECREF on the job wrapper in PyEventLoop::enqueue and _enqueueWithDelay, plus a regression test.

Why

The wrapper is passed to call_soon_threadsafe / call_later with the "O" format, so the returned Handle / TimerHandle takes its own reference and holds the wrapper until the job runs. The wrapper's creation reference was never released, so the wrapper — and the JS job closure it owns via m_self — leaked once per scheduled promise job and accumulated for the life of the interpreter. The added Py_DECREF drops only that creation reference; the handle's own reference keeps the wrapper alive until the job runs, and the wrapper's normal teardown releases jobFn.

Reproduction

I compiled a gist with reproduction details which should demonstrate the leak.

I ran into this when converting spreadsheets to JSON using a library which unzips the .xlsx file using an asynchronous library and I saw the used memory grow by about 200-300MB with every file I processed.

P.S.

Thanks for an amazing library, I hope I didn't misunderstand the underlying issue, and I'd be happy to change the approach if necessary.

[steinnes]

I've tested this locally on my mac, as well as using docker builds, but I haven't tested it on Windows. I looked around for guidelines on contributing but didn't spot any, hopefully I didn't miss them and make any obvious mistakes which waste your time 🙏