My 2 cents. As @colinbrislawn mentioned I think you can choose either of those cases because they produced negligible (in my opinion) differences. I’m guessing the bulk, if not all, of the increase in your features # probably comes from singletons or features with very low frequencies in a handful of samples only. You can check this for yourself. These are typically removed when you are doing differential abundances analyses anyways as they don’t really provide much information and increase noise, so it won’t affect anything there, and in indexes where they drive alpha-diversity changes see this great discussion about why perhaps those are incorrect to begin with.
Ultimately you can look at the sequencing depth of your samples between the two cases and see if one allows you to retain more samples. In that case, I would stick with that option.
Edit: After re-reading the discussion I linked myself, I realize perhaps most of those feature increase are not singletons per se, as DADA2 tries to avoid those, but I think they would still fall within the realm of very low frequencies within very few samples.