Setup CPU profiling and optimize code

[arostamianfar]

We have not really spent too much time optimizing our code, but it seems like there may be opportunities to gain performance. For instance, a small change in PR #417 resulted in 50% speedup in one of the PTransforms (~10% overall speedup; or more depending on the size of the data).

However, rather than trying 'random' things, it's better to run a CPU profiler on our code to determine the places where it makes sense to optimize (e.g. I have found that sanitizing the BQ fields also takes a nontrivial amount of time). Beam has some flags to setup profiling (see thread).

We can also consider using Cython once we know the areas to optimize.