Automatically drop NaN values when executing ancom.py

[valentynbez]

During the analysis, I had some missing metadata values. To counter it I had to:

• revert to a feature-table step
• filter it, filter than collapse it
• add_pseudocount
• filter metadata
• finally run ancom

Missing values in metadata are human-introduced errors, they appear quite often. This is an inconvenient workflow for the user.

Instead of raising an error when metadata has nan entries, I propose ANCOM would proceed and inform the user which samples were excluded from analysis if any.

[ebolyen]

Hi @valentynbez,

Thank you so much for this, and we're really sorry we didn't get back to you sooner.

We'd ideally handle this via an initial error with a new parameter which could override the error, instead of always allowing it through with a log entry.

The reason for that is a user may not realize they have missing entries, so the positive affirmation to filter is a useful sanity check for them in the event they didn't realize.

I realize this is so late a response that it's unfair to expect any changes at this point, so I'm going to create an issue and reference this PR as an initial implementation for your feature.

[ebolyen]

Closing this for now in favor of #90 because we're going to rework some of the details, but let me know if you'd like me to re-open this if you're still interested.