Hi @rahel_park,
Once again, apologies for the late reply - and thanks so much for your patience! Happy to address your questions below:
We chose not to support continuous variables because this can diminish the ability to make meaningful comparisons between groups. Variables with discrete numerical groupings are still appropriate to use, you would just need to identify that within your metadata as a categorical column using the comment directive.
Multiple (3+ groups) can be used within the QIIME 2 implementation of ancombc because the features mentioned in the warning you received (global/pairwise/dunnet/trend) are not implemented. It is interesting though - to my knowledge, those additional tests (pairwise, dunnet, trend) were only available within ancombc2.
For full disclosure, I do not have as deep a familiarity with ancombc2 as I do with ancombc - this is a method that we will be implementing into QIIME 2 this year, but it is a newer method that our core development team are still actively learning about for development.
It looks like @jwdebelius did a great job of addressing this question in your other post here, so I'll leave this as-is to help keep threads tidy 
As I mentioned above, I can't speak with as much confidence regarding ancombc2 - however, I do wonder if this is because you've chosen a continuous variable in your formula. This is essentially the reason we've removed the ability to include continuous variables within the QIIME 2 implementation of ancombc. It can be tough to group continuous variables (since essentially everyone is in their own 'group') - which can produce groupings or significance that aren't really meaningful.
Hope this helps! Cheers 