Logging of cpu-seconds will miss subprocesses that exit before the main process

[berland]

The polling nature of cpu-seconds counting in

ert/src/_ert/forward_model_runner/forward_model_step.py

Line 206 in 7c27a9c

(memory_rss, cpu_seconds, oom_score, pids) = _get_processtree_data(process)

will underestimate the cpu_seconds if the main process for the forward model step launches subprocesses that end earlier than the main process, as the process tree for a process that has ended does not exist in the kernel any longer.

[berland]

It is probable that this happens for the RMS step, as it is known to launch subprocesses for parts of its doing.