As well as posting a complete example that demonstrates a difference between 9.2.8 and an earlier version, you might like to try :i SomeConstructor in GHCi. I think it’s quite likely it’s picked up a type argument that wasn’t there before (perhaps because of PolyKinds?).
type AConstantVia :: forall {k}. k -> Type -> Type
newtype AConstantVia mods a = AConstantVia {unAConstant :: a}
a = AConstantVia @('DropLeft 3) FooBarBaz
By marking the k type variable as inferred, it’s now hidden from type application.
I think the right way forward is to be provide a kind signature for every datatype for which type-application is expected. Otherwise the inferred kind could change between the GHC versions, but you also need to guess the correct parameterisation. I also think they make more complicated types easier to parse and you improve readability. I got quickly used to them.
I also do think we should have this documented somewhere. I know ours is a bit of an esoteric use-case, but other people might also stumble upon the same issue.
The GHC migration guides might have coverage on this issue. I’m not sure what version past 8.6.5 introduced the breakage, but it’s probably worth skimming over the guides to see if it’s covered already. Going from 8.6 to 9.2 is a pretty big jump!