Good catch! Yes, truncation happens at the end of the read, which is on the right for these graphs.
Here's my best write-up on how to calculate the expected overlap for an amplicon:
The truncation length of forward and reverse do not have to match.
(Because the quality of R2 is often lower than R1, people often trim more from R2.)
Yeah, try those!
If you go too short then the percentage of input merged will drop after running DADA2. You can see this in the stats.