pps: do not change scheduler in GpuDataSource::trace

Calling sched_setscheduler twice on each trace call has a high cpu
overhead.  intel and panfrost use HW/kernel-based sampling.  freedreno
has switched to pps_rt_sampler.  We do not need it to avoid preemption
269 jobs for !19426 with pps-overhead in 7 seconds (queued for 8 seconds)
latest merge request