Fix worker death

[josephjclark]

Short Description

This PR fixes an issue where the worker doesn't close down pooled child processes after uncaught exceptions, resulting in the worker refusing to work.

Related issue

Fixes #664

Implementation Details

What's basically happening is:

• The engine is correctly catching uncaught exceptions (and I think process.exits) when they occur
• And it is correctly sending error events out (these errors may be rubbish but I do think they are sending)
• BUT the engine doesn't properly resolve the promise for that run, so the engine still thinks the workflow is executing
• That means the child process pool doesn't allocate new resources for future work, and so chokes up

The engine should eventually be timing out the runs, but by this point lightning thinks they're dead and isn't listening to events. But I think the backlog will very slowly clear.

Anyway, as result of the fix, the error is handled gracefully, the pool re-allocates the worker thread, and everyone is happy.

QA Notes

I've added two integration tests, both of which reproduce very similar errors to main. And they both fail on main.

Checklist before requesting a review

• I have performed a self-review of my code
• I have added unit tests
• Changesets have been added (if there are production code changes)

[josephjclark]

This is incidental debugging code. I do want to leave it here but commented out for now.

[josephjclark]

This is the actual fix and this is what WASN'T happening before.

It's not enough to tell the worker we've errored. We need to send an event to the parent process to tell it that we're dead.

I'm a little concerned about duplicating error messages here and so there's a little bit of gnarly code on the reporting side to handle that.