Add threshold options to kill child if goes awry

Inspired by
- https://github.com/peak/s5cmd/issues/746

where child process can really get out of hands and consume too much, causing linux to start killing other processes instead of the abuser (unlikely but possible). Since we are monitoring anyways, I wonder if we should add facility to kill child process (and kids) upon
- exceeding memory, cpu, wallclock time thresholds?

somewhat relates (likely in implementation):
- https://github.com/con/duct/issues/184

but also likely there is already a tool for that, so we might just want to "pair it up" one way or another.