Improve internal observability via catching a special signal and dumping in-flight work

[sdboyer]

At the moment, it can be quite difficult to tell exactly what work dep is doing when it hangs (or at any other time, really). There are various strategies we might take to improving this in the longer term with our overall output, but a short-term strategy that could yield major benefits would be to add a signal handler (so, non-windows) for, say, SIGUSR1 that causes the SourceManager to dump out a list of all the supervised processes.

It should be possible to do this relatively simply with a single new method on SourceManager, then a signal handler installed up in dep to call that method.