Let's see if we can figure this out...
I assume you mean that you downloaded a pre-curated set of files from the QIIME 2 website?
If you are using SILVA 138.1, then you should be fine. Nothing should have changed within the SILVA database files between QIIME2 versions.
There is nothing wrong with the SILVA processing commands you used. However, there are some other steps that were carried out on the premade QIIME 2 files. These differences could affect how things are classified. I think all you need to do is run cull-seqs on your extracted reads to clean things up a bit.
This is fine and intended. I strongly suggest that you leave these reference sequences within your database! Many 16S rRNA sequences can amplify eukaryotic 16S rRNA sequences. It is best to be able to identify them, if they exist, so that you can remove them. If not, you run the risk of erroneously classifying them as Bacteria, when they in fact are eukaryotes. Reference sequence / taxonomy classifiers should always contain outgroup taxa.
Again, it is recommended that you do not do this. Also, there is no need to write your own scripts. You can simply run qiime taxa filter-seqs ...
It looks like you are running an older version of 2024.5 which had a bug. I'd re-install this version... or better yet, simply install 2025.4.
Again, I'd install the latest version of QIIME 2 to make sure there are no other environment errors. Also, can you share the QZV of the assigned taxonomy for both runs? I'd like to look at the provenance.
Also, The later version of RESCRIPt now use SILVA v138.2 by default. Nothing is much different other than updated taxonomy labels.
I am not sure what you are trying to say here. Are you referring to classified features in your data, or reference features for your classifier?
-Cheers!