Add basic provenance tracking

[tavinathanson]

Since I'm regenerating a shared cache right now, I figured I'd knock out some basic provenance tracking this time around.

What this does:

• Records the versions of a few key modules alongside each cached file.
• If those versions don't match up with the environment when loading from the cache, simply prints the diff for each patient.

Future work:

• Allow multiple cached versions to co-exist, keyed by the provenance. Currently, this is just an alert rather than a useful tool.
• Somehow track NetMHC* versions.
• Use logging rather than print statements and have a way to suppress and/or merge the many warnings.

[tavinathanson]

@armish I'm writing up a small test right now, and decided I should probably disable the warnings by default.

[armish]

how about something like:

def check_diff(diff):
  if diff != set():
    print("Current provenance for patient %s: %s" % (
      patient_id, provenance_str_from_set(diff)))

check_diff(new_provenance_diff)
check_diff(old_provenance_diff)

to reduce the redundancy here?

[tavinathanson]

I reduced the redundancy somewhat differently, let me know if it looks good?

[armish]

nice one! 👏

Just a few minor questions/comments for you. Feel free to merge it when/if you address/ignore them.

[tavinathanson]

Updates:

• Added a test.
• Added a check_provenance flag that defaults to False.

[tavinathanson]

@armish Good to merge on green?

[tavinathanson]

Fixes #15

[armish]

Go for it!

BTW: I think this looks much better with df -> dict switch. Thanks for doing that.

[coveralls]

Coverage increased (+2.8%) to 59.267% when pulling 526030f on provenance into cb01071 on master.