scx_lavd: run_time_ns may vary too many for Dhrystones like process

[archerbroler]

Dhrystones like process is a single thread with high acc_run_time_ns accumalated.
acc_run_time_ns will be set to 0 when runnable, but thread like this will keep running resulting in larger than 1s acc_run_time_ns .
When thread like this quiescent and then runnable, the acc_run_time_ns is set to 0.

run_time_ns is update when stopping, this means after a long single run, run_time_ns is avg calc with the new much smaller acc_run_time_ns. Which will cause a big drop in run_time_ns value.

This influence the perf_cri, lat_cri, vdeadline and est_stopping_time calc.

I think the EWMA are weak to situation like this. Maybe util_est in PELT like method can help?