Allow killing processes with SIGTERM

[lionel-]

Branched from #357

• Add cleanup_signal argument to run() and process$new(). This instructs processx to use softer termination on GC so that the callr cleanup handler may run.

• Add signal arguments to kill() and kill_tree() methods. Used internally to support the above, and externally by event loops.

• Since the kill signal may now be ignored by the process, poll for SIGCHLD with a timeout. This is implemented using process_wait() and tested with a new dynlib sigtermignore that installs a SIGTERM handler that ignores the signal.

I'm not 100% happy with the API implemented here. I can call kill_tree(signal = SIGTERM) in an event loop and register a subsequent SIGKILL with a delay, this part is fine. However, when processes created with cleanup_signal = SIGTERM fail to properly quit on GC, the processes might not be killed. Also it seems that close_connections = TRUE also causes the process to finish (e.g. it causes processes linked to sigtermignore to terminate), and I'm not sure how this interacts with SIGTERM cleanup.

Would it make sense to revive the grace argument and do staged termination with SIGTERM and SIGKILL by default?

[lionel-]

Would it make sense to revive the grace argument and do staged termination with SIGTERM and SIGKILL by default?

Done in #367 and r-lib/ps#149.