For documentation purposes, I think your models are fine. As far as I know, a message flow connected to a collapsed sub process does not have the semantics of instantiating the subprocess. It merely means that in the context of the sub process, a message is received. At least that is the case for collapsed pools.
As a rule of thumb, try to model it the way you think the people reading the model will understand it. If you believe they will be unsure about the semantics of the second model, then it may be a good idea to improve the model by a) adding documentation elements explaining this b) modeling this in another way or c) leaving the model out.
For automation purposes, it is not really a problem either, since Camunda does not interpret message flows.