Get fill_prob_matrix tests passing

[glesica]

Right now we're attempting to verify our results against those from the Perl script. There are two outstanding challenges, however:

1. The numbers are "off", but this may be explained by the fact that the Perl script was using a 0.0 change probability. We may want to fix the Perl script, re-run, and then check the tests with the new results.
2. The Python script reports row / column results that are omitted from the Perl output. We should decide how we want this to work and harmonize the two implementations. We probably want the Python script to work like the Perl script.

[glesica]

The description above is no longer accurate, but we still probably need a couple tests for this function.