crypto: msm: optimize ce bandwith scaling

Encryption processing at qcrypto driver includes three main steps:
set ce high bandwidth and enable ce clock in cra_init, then send
encrytion request, and at last set ce low bandwidth and disable
clock in cra_exit. This process is fine for the use case where
there are lots of requests between cra_init and cra_exit, but will
introduce large overhead for mmc layer disk encryption as it calls
cra_init, queue request and cra_exit for every encryption request.

It is not necessary to set low bandwidth and disable clock for every
encrypt request when it is completed. So we delay these scaling down
tasks for a while, and just scale down the last request when timer
expires so as to amortize and decrease the overhead.

Change-Id: I6a59066343f01950b66f4e886d40a9f27ce211c3
Signed-off-by: Zhen Kong <zkong@codeaurora.org>