Re-assess GSEA results at tumor purity threshold

[sjspielman]

Closes #1626

This PR updates the GSEA module to assess to what extent key results (indicated in issue #1626) might change if we perform GSEA only on tumors that pass a purity threshold.

Punchline: Results are almost entirely consistent ✅

There are several key code changes:

• Added an flag in the module crux script to generate results from thresholded data
• Added a notebook to process these results and compare them back to the original results. This notebook is quite granular and walks through each comparison we emphasize in the manuscript. While there is some (not surprising) loss of power in certain comparisons, broadly speaking results are consistent! This notebook can therefore serve as a formal record of the result consistency. Here's the rendered version: 03-assess-gsea-at-threshold.nb.html.zip
• Both of the above are now in the module run script and documented in README which was amusingly out of date
• There are various HTML diffs to the 2nd notebook from re-running (I wonder if it was not originally generated in docker, which would have been my bad!)
• Some tiny recreational revisions......
  • I removed a print statement
  • Updated the significance threshold in the 02 notebook to be 0.05 which is what we use in the MS, but results presented in the notebooks do not change! The result files got slightly tweaked because they're binary, and the TUKEY results have a column indicating significance so that column may be a tad different. But, it's now more consistent with the MS which I think is probably slightly better.

[sjspielman]

I addressed reviewer comments including:

• Not quoting things that aren't quotes, given that there are quotes!
• Doc updates and argument name updates for clarity
• Comparisons across a given set of pathways, not pathways themselves, are up/downregulated.

In https://github.com/AlexsLemonade/OpenPBTA-analysis/pull/1659/commits/39cdbc132832b63bc0f8a844dacc36c9402973e8, I incorporated the tumor purity filtering function and re-ran the module. There were no diffs from re-running.