Clean up some benchmark code and docs.

[asfimport]

Strictly following the instructions on the parquet-benchmarks module doesn't give meaningful results.

It appears some new benchmarks enter into conflict with the globs specified for others, not all benchmarks are run, and some iterations of write benchmarks aren't evalulated due to unexpected "file already exists ..." fail-fast returns in the data generator.

This should be cleaned up to encourage using and implementing benchmarks on Parquet code.

Reporter: Ryan Skraba / @RyanSkraba
Assignee: Ryan Skraba / @RyanSkraba

PRs and other links:

• GitHub Pull Request #672

_{Note: This issue was originally created as PARQUET-1644. Please see the migration documentation for further details.}