Add docs, unit tests, and refactor repro test infra

[charleskawczynski]

This PR adds docs, unit tests, and refactors the reproducibility test infrastructure. In adding unit tests and documentation, I've realized that some of the terminology around the repro tests needs revised for certain terminology to be well defined.

In particular, if we consider the situation where commits are reverted, we can end up with multiple and diverging paths (multiple paths with the same reference counter, but made at different periods in development). For example, if our reference counter history looks like (top being most recent):

137
     /PR10 # is PR6 even valid to compare against, here?
136
     /PR9
     /PR10 # should this compare with PRs 4 and 5?
135  # Reverted
     /PR8 # should this compare with commits in PRs 1,2,3?
138
     /PR7
137
     /PR6
136
     /PR4
     /PR5
135
     /PR1
     /PR2
     /PR3

We could clean up older commits, but I think a safer option is to change the latest_comparable_paths to use the (newly) added function compute_bins. I originally used the term bundle (e.g., compute_bundles), which I'm almost leaning towards returning to. It's a longer name but sounds more...tidy.

This was previously not an issue when latest_comparable_paths was latest_comparable_path, because we always grabbed the most recent matched reference to compare against, eliminating this potential issue.

I'll update the README once we land on a design that is coherent, but I think it might take a bit more refactoring to get there.

[Sbozzolo]

Thanks! I think this is a big improvement in transparency on what the reproducibility tests are doing.

I just left one comment for a function whole logic I have hard time following