Don't SIGKILL my job bro!

[jasonmads]

SIGKILL (9) is harsh
Using SIGKILL to kill my job doesn't let it shutdown gracefully. My particular application will write a checkpoint and exit for a SIGTERM. Additionally, if you have a wrapper that launched a subprocess, you can (I do) end up with an orphaned subprocess.

I have a few ideas
Not 100% what would be the best way...

1. Just change KILL_SIGNAL to SIGTERM, not a good idea.
2. Making KILL_SIGNAL layer option defaulting to what is set in rqconstants.py at runtime might be the lest disruptive, forgoing the term then kill approach. DB/cuebot/proto/rqd edits
3. Have rqd check for layer flag and kill based on that. less visible, but only rqd edits
4. Expose KILL_SIGNAL to rqd overrides, it become a per machine thing. just a little rqd edits
5. SIGTERM, timeout to SIGKILL approach, die or I kill you approach. What is the timeout? overrides?