• Adds support for modifying child process oom_score_adj values. The logic for the child's score is:
  • If agent score < 0 then 0
  • If 0 <= agent score < 998 then 998
  • If agent score >= 998 then 1000
• Reduces potential for log spam when we're not permitted to adjust a process's niceness or oom_score
• Adds a CODER_PRO_OOM_SCORE for specifying an explicit value for child oom scores

It's still the responsibility of the template admin to run the agent with a customer oom score since it's awkward for us to set it ourselves (we need cap_sys_resource and it needs to be set on the binary since we can't elevate or own privileges if we're ever run as non-root)...if people want this we can eventually support it.

@deansheather I didn't go with the solution we talked about since I was encountering some inconsistencies when adding a coder agent exec command...I think it's due to needing to lock the OS thread we're running on but I had to do some jank to make it work so let's hold off until this solution isn't sufficient.