Make parallelism optional

[dan-mccabe]

Originally raised by @robfitzgerald:

Feature

multiprocessing is currently being used to parallelize operations, but these operations may actually perform more slowly than synchronous code, due to the fact that:

• we need to slice up dataframes and copy them
• multiprocessing brings it's own runtime performance hits for forking and joining (aka, pool and map) processes

in order to review the performance benefits, we can create a separate workflow based on the presence of a "parallelism" argument where:

• parallelism is None or 1: skip parallelism, do work directly on dataframe
• parallelism greater than 1: use current workflow

Analysis

with these changes, we should run a test for parallelism values to discover whether runtimes are improved.