Sorry, culpa mea, I wasn't reading closely enough the first time around. Got it now.
That's easy. You can filter a sequence artifact to match only features found in table X by using feature-table filter-seqs. Just do something like:
qiime feature-table filter-seqs \
--i-sequences sequences.qza \
--i-table filtered-table.qza \
--o-filtered-sequences filtered-seqs.qza
That should do the trick.
In theory, no you do not really need to filter seqs. For all downstream statistical analyses that I can think of, you don't even use the sequences file itself so having extra seqs does not really matter.
Where it does matter is for steps that do take a sequence file: taxonomy classification (which you've already done), alignment/tree building (which you've probably already done, too), and some other non-core analyses (like some methods in q2-quality-control). Having these extra sequences will just slow down those steps, that's all. So probably not important, but sometimes it's best to cut out dead wood and not have to worry about the details (at least when the method exists and is easy).
I hope that helps!