simplify docker failure code

I think we should simplify the docker failure scenarios.

a: If docker is requested and we can't run docker, then fail.
b: if a dockerfile is supplied but can't be used, then fail.
c: if docker is requested but no dockerfile, then use the built-in one.