sched: core: Fix usage of cpu core group mask

To find number of clusters, we depend on topology_core_cpumask(),
but that returns current online sibling cpus, if cluster have
offlined cpus, then finding no. of clusters could go wrong.

If system booted with less no. of cpus and others comes online after
bootup, then sched_up/down migrate setup breaks.

Fix this by using core's possible sibling mask.

While at it, refactor cpu possible siblings mask API to avoid other
architecture compilation errors.

Change-Id: I4543ad5b711181e60271fa887bb7d5855a55ba90
Signed-off-by: Lingutla Chandrasekhar <clingutla@codeaurora.org>