After some searches in the forum here, I processed my forward and reverse data separately using dada2 denoise-pyro. and then applied feature-table merge with --p-overlap-method sum for the tables and feature-table merge-seqs for the ref-seq files.
I think it was the best option to handle the data here.