add compositionally aware paired test

[nbokulich]

Improvement Description
Add a compositionally aware method for paired testing of a feature table

Current Behavior
pairwise-differences accepts a feature table but the tests implemented (paired wilcoxon, paired t-test) are not really appropriate for relative frequency data.

Proposed Behavior
I see two options:

1. implement a new action specifically for feature tables that performs an appropriate test. E.g., a paired ANCOM test would be quite simple to rig up. A new visualization would be nice 😄 though the old one would work, with modification for multiple testing.
2. adjust pairwise-differences to accept only transformed feature tables, and implement an appropriate transform somewhere else (e.g., ILR might be implemented in q2-classo?)

References
forum xref