You are absolutely right on the filtering aspect. I wasn't taking into account that 1) the F & R reads were very low quality greater than 200 nt in length and 2) since I used cutadapt to remove primers and spacers prior to DADA2, my F reads were now only ~220 nt and my R reads ~210 nt, so with my truncate length in the original script of F @ 245 and R @ 220, I was losing almost all reads at that point anyways since the reads after cutadapt were already shorter than my truncate length.
I re-ran everything with the DADA2 forward and reverse truncate length at 200 and 180 nt respectively, and got what seem like more reasonable results. After filtering, denoising, and chimera removal, I am left with ~40-60% of sequences for most samples. I would hope for more retention, but I may have to play with the DADA2 parameters a little more.
I am still a little concerned about orienting the overlap based on V4 size (250 nt) versus amplicon size (350 nt). My thought is, and this is just a guess, we should take into account the full amplicon size, because we are sequencing the entire amplicon, not just the V4 region, correct? so if we truncate both reads too short, then there won't be any overlap. It looks like @Mehrbod_Estaki summed it up well in this post where he said to use the amplicon size: Low frequency counts after DADA2 (7%) - #3 by Mehrbod_Estaki.