scx_layered: Improve yield, preemption and other behaviors

An yielder is now charged the full slice by default.
The amount of slice which is given up on each yield can now by modulated per-layer using the new yield_ignore parameter.
Preemption is now first tried on the previous CPU the preempting task was on which should result in better locality behavior.
Confined layers can now preempt.
A new layer property preempt_first is added which prefers to preempt the CPU the task was previously on over idle CPUs.
Long running tasks now can keep running on the same CPU within max_exec_us.
More behavior improvements and statistics.

sched-ext / scx