Hi @sdpapet,
Great questions! I will answer specifics below, but in general I recommend that you see the overview tutorial, which will clarify many of these questions.
No. See the overview tutorial but, better yet, the dada2 paper and this paper to answer all of your questions in question 1. There are many more papers out there on this topic (both benchmarks and reviews) so I recommend a literature search for more information.
All of those methods are available in QIIME 2 and covered in this tutorial.
There is no "default" in QIIME 2... QIIME 2 is a platform containing many different methods. A sort of "choose your own adventure". We do recommend and prefer denoising methods over OTU clustering, and the resources I recommend above will explain why.
Because denoised sequences do not require clustering. See the resources above for more details.
None of the tutorials use BLAST against NCBI for actual taxonomy classification. This is a feature of qiime feature-table summarize, which allows quick lookup of specific features, and is provided for user convenience, NOT for actual taxonomic classification.
Read further down in the tutorials, e.g., here, to see how taxonomy classification is performed. Several classification methods are supported in QIIME 2, and any of those reference databases that you described are usable — we even have pre-trained classifiers and links to compatible reference databases on the data resources page.
Taxonomy information is not stored in feature tables, because doing so makes it ambiguous whether any given feature table is annotated or not with feature metadata. Any method in QIIME 2 that requires taxonomy information (or other feature metadata) requires that information as a separate input. It makes things much more explicit and clear compared to, e.g., qiime1.
See this tutorial to learn how to append this information if you are exporting your feature table for use with other software.
Good luck! I hope that helps.