scsi: ufs: update clock scaling sequence

According to UFS host controller hardware program guide, driver should
always issue hibern8 enter before start scaling clocks, regardless of auto
hibern8 is enabled or not. Meanwhile, in current clock scaling sequence, if
scaling up/down clocks fails, driver bails out with error but leaves the
link in hibern8 state. In this situation, subsequent tasks can not be
served, thus task abort shall happen. This change fixes it by issuing
hibern8 exit in the clock scale error out path.

Change-Id: I32d8f9d5f313fa7a566de1d90df5004ad20a3d09
Signed-off-by: Can Guo <cang@codeaurora.org>