mtd: msm_qpic_nand: fix a race condition in page read

During a recent performance optimization patch (commit sha:
09b431ab, commit summary:
"mtd: msm_qpic_nand: reduce the number of ce"), there was race
condition introduced for a page read opeartion. If there are
more than one client for a NAND read, then its possible to hit
this bug.

During the race condition, the READ_LOCATION_0 is not written and this
leads to reading an incorrect number of bytes per codeword sometimes. Fix
this protecting the code with a mutex.

CRs-Fixed: 650682
Change-Id: I58426a2b9005619901173c3e87da7169024e5041
Signed-off-by: Krishna Konda <kkonda@codeaurora.org>