sched: optimize select_best_cpu() with same power cost CPU mask

At present, the function select_best_cpu() has to continue to iterate the
search CPUs for the possibility of finding a more power efficient CPU even
if the best CPU based on cstate already found.  However many of MSM
big/Little processors share an identical power cost information table for
the power efficient cluster CPUs so we can avoid this unnecessary searches
mostly.

Introduce kernel parameter 'sched_same_power_cost_cpus' to inform the
CPU mask that has an identical power costs and optimize the 'for' loop of
select_best_cpu() by rulling out CPUs that cannot be more power efficient
than cstate based best CPU with the same power cost CPU mask.

CRs-fixed: 849655
Change-Id: I9d2e7f560282a914fd0373aac6714353b9d1e36c
[joonwoop@codeaurora.org: as this branch doesn't practically support EA,
 trimmed out all about kernel parameter 'sched_same_power_cost_cpus' but
 adopted idea to clear out search CPUs upon min_cstate == 0.]
Signed-off-by: Joonwoo Park <joonwoop@codeaurora.org>