Yes and yes. I cut scenes up if they're long and I put the extracted faces for Part A and Part B in the training thing together, train everything, then I put just the faces for Part A and the Part A frames in data_dst, run training for just a bit with no warp, the final sharpening stuff, then merge.
Then replace the faces with the Part B faces, put the Part B frames in, train for a little bit (same thing) and merge that. You could do the same thing with a different clip, it'll take a bit of training to look great but it'll go way faster.