yes this is supported. This should be in the documentation here:
Thanks for the link, but just wanted to make sure we are speaking about the same things.
After head swap new controller most likely has different NVRAM size so NVRAM mirroring is disabled. Actually for 4-node cluster with ARL head swap MC FC FAQ explicitly warns "NVRAM mirroring between the sites must be disabled for the entire process" which logically also applies to 2-node cluster. And it goes on to add "Planned switchover is not possible while the upgrade is in process".
But for non-disruptive 2-node head swap we must do planned switchover (and switchback). It is clear that initial switchover with old heads is possible. But the main question is - will switchback and subsequent switchover in opposite direction with one old and one new head be possible too? If yes, is it done by standard commands or some diagnostic mode magic is necessary?
Documentation rather sidesteps it unfortunately ...