Hello,
As far as I understand, currently Meshroom chooses origin based on the initial pair of images (at the "StructureFromMotion" stage), so it's basically random. This leads to grid and model being usually rotated at some unnatural angle.
This obviously can be easily fixed in any 3D editor afterwards, but I think it could be easily done automatically by analyzing all camera poses - their local Y-axes should be usually pointed upwards. When "up" direction is determined, Y coordinate can be selected as a minimum of all points, and X/Z as their average.
It's a simple feature, but it would improve visual preview in Meshroom and probably remove the need for one manual step in post processing in most cases.