What I would recommend doing—and I know this is going to require more work—is to start with the probably more clearly desirable transformation, i.e. merging the structure fields together. That could even be done one patch at a time, i.e. merge u.audio_mpeg1 into u.audio, then u.audio_mpeg4 into u.audio, etc. Then, after that, we can consider turning major types into minor types.
I'm agree with that, it's a good suggestion. I'm also struggled by splitting this patch.
I'll open a new MR which merge audio_mpeg1 into u.audio to see if that's acceptable.