For the new result objects, do they override checkpoint/early_stopping metric names set by trainer callbacks?
I believe so - see here.
In theory, you could override this implementation in a subclass to use whatever name you want for the early stopping metric / checkpointing metric.