soc: ti: knav_qmss: Use percpu instead atomic for stats counter

soc: ti: knav_qmss: Use percpu instead atomic for stats counter

Hwqueue has collect statistics in heavy use queue_pop/queu_push functions
for cache efficiency and make push/pop faster use percpu variables.
For performance reasons, driver should keep descriptor in software handler
as short as possible and quickly return it back to hardware queue.
Descriptors coming into driver from hardware after pop and return back
by push to reduce descriptor lifetime in driver collect statistics on percpu.

Signed-off-by: Vasyl Gomonovych <gomonovych@gmail.com>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>