Hi JoseASanchez,
yep, managed to build an orthomosaic but completely abandoning the two camera approach. Using the calibrated RJPG (now .tiff files), I have stitched them independently by subset them in three different chunks ( I have in tot 800 images). Fore some reason, not completely clear to me, when the images are divided into chunks metashape is able to align them (while processing all the images together fails). The alignment process is complete after iteratively running the alignment several time (as in
https://www.agisoft.com/forum/index.php?topic=8554.0 ) and adding GCP for each chunk. I have subsequently aligned the three chunks using GCP (marker). Consider that in the calibrated .tiff are written, using exiftool, the original metadata of the rJPG about centroid coordinates, altitude and yaw, pitch, roll degree. I guess without these metadata the alignment would fail more likely (but depending on your images resolution..)
Happy to give further details if needed.
Matteo