It's a nice idea, and we have experimented in the past with something similar - deriving headings from a pair of low cost (L1) RTK units, with surprisingly good results - see
this post for more.
Actually combining two separate positions on-the-fly to calculate headings would, I suspect, be a rather niche requirement so I am not sure it would be a top priority for us. Better perhaps to log the raw data and post-process, before importing into RM? We have synchronisation of independently logged GPS data with sonar files as an upcoming feature later this year, which can really improve sidescan mosaic positioning.