OpTracker: Sharding logic is implemented to improve performance

OpTracker: Sharding logic is implemented to improve performance

Enabling OpTracking is degrading performance significantly. The
main reason for this is the contention in the ops_in_flight_lock
during registration/unregistration. Sharding this lock and hence
the xlist (storing the ops) helped to boost the performance.
Sharding is been done based on the op seq number % num_shards.
atomic64_t is used as data type to protect the global sequence
number of OpTracker. Introduced a lock in the OpHistory to
protect its data structure as ealier it was protected by global
ops_in_flight_lock. This is only accessed during unregistration
and thus not contended.

Signed-off-by: Somnath Roy <somnath.roy@sandisk.com>