Test std Mutex vs parking_lot

[arriqaaq]

          Seems like std Mutex performs better when highly contended which is our case with a large number of concurrent transactions. I've updated the PR to use parking_lot just for the sake of uniformity with other locks in the code base, but sometime later we might want to change every Mutex and RwLock to std and run the benches.

Originally posted by @gsserge in #166 (comment)