Good to see you again Devon. Let's figure out the read cutting settings and cut to the feeling. 
You have already identified the core issue:
By default, the dada2 function mergePairs() includes these settings
minOverlap = 12, maxMismatch=0
so no wonder many of your reads aren't merging!
Frustratingly, the dada2 qiime plugin doesn't expose many settings that would let us make the read joining more permissive, but fortunately vsearch does. You could increase fastq_maxdiffs even more to get even more reads to join.
Given that dada2 isn't working on your reads, I think you could proceed forward using a different clustering denoising method. Specifically, you could try deblur or classical OTU clustering, both of which will run fine on paired end reads joined with vsearch.
The results of these will be a feature table with rep sequences that you can use for taxonomy classification.
Hopefully this get's you off to the races in the new year! 
Colin