ANDROID: cpufreq: Improve util tracking and add pressure

There are VM usecases where physical CPU frequencies can exceed
the max allowed frequency given to the guest, relax the requirement
for freq_scale such that:

freq_scale * arch_scale < sched_scale_capacity^2

This guarantees pelt_clock is less than elapsed time, and will allow
for more accurate util accumulation and tracking for tasks inside vCPUs.

Add thermal pressure when userspace limits scaling_freq_max for
better task placement decisions inside the guest. Thermal pressure
is reworked as generic system pressure in the link below and will
supersede this change.

Bug: 284368385
Link: https://lore.kernel.org/all/20240326091616.3696851-1-vincent.guittot@linaro.org/


Change-Id: Ibdefc41b9ced3d4686cbc6b802270716621be944
Signed-off-by: David Dai <davidai@google.com>