I too was concerned about junk sequences, but we have not done a good comparison with Miseq (or some other platform) yet. Note that Singer et al. 2019 did such a comparison and suggest that the NovaSeq is just plain better: Comprehensive biodiversity analysis via ultra-deep patterned flow cell technology: a case study of eDNA metabarcoding seawater | Scientific Reports
We have used the 515-806 for 16s and its1f-its2 for its1.
In some of our preliminary work using vsearch to generate OTU tables, we have made truly monstrous tables (Tb size) that are impractical. We are considering hdf5 and other tools to try and reduce memory footprint. If anyone has suggestions I would love to hear them. Similarly, I am curious regarding compute time for dada2 operations for such large data. I haven't had a chance to do any benchmarking myself yet.