adonis: add option to ignore NaNs

[nbokulich]

Improvement Description
Add an option to ignore NaNs in adonis. This would be similar to the --p-ignore-missing-samples options (with various names) in q2-emperor, q2-sample-classifier, probably also q2-longitudinal...

Current Behavior
Since #303 , a nice error is raised if NaNs are detected in the input data. Pre-303, a not-nice error (from R) was raised.

The only way to bypass this is to filter samples from the metadata and feature table manually, then repeat.

Proposed Behavior
Add an option to instead filter the table and metadata by dropping samples with NaN metadata values, instead of always raising an error.

1. This parameter should be False by default (i.e., raise an error by default instead of dropping samples)
2. The number of total and dropped samples should be reported in stdout
3. The number of total and dropped samples could be reported in the adonis visualizer (e.g., similar to what other visualizers in q2-diversity and q2-longitudinal do)