Improve laziness of typeclass derivation

[ghostdogpr]

Continuation of #594 and @darl's work to make sure Magnolia uses the custom typeclass instances before it tries to derive it.

[ghostdogpr]

There is one drawback to having one level of indirection when calling Magnolia, which is that you don't get Magnolia detailed error messages when calling gen[A] explicitly. Instead, you get an error saying it can't find the implicit Derived.

You can get the detailed error by calling autogen[A].schema but that's less straightforward.

[ghostdogpr]

Maybe it's fine if I document it. gen[A] will still work so no breaking change, and autogen[A].schema can be recommended for debugging.

@paulpdaniels any opinion on this?

[paulpdaniels]

I'm fine with this, it's all still magic to me. I wonder though could we have some way of surfacing the autogen distinction in the error message? Or perhaps we could add it to the scaladocs on gen?

[ghostdogpr]

I don't think we can surface the error message because gen is not calling the macro itself, it is only checking if there is an implicit in scope.

Good point on the documentation. I think I will rename autogen to genMacro too, the difference will be clearer.