Allow buffers to specify their message type

In this PR conversation I brought up the idea of having an optional "stable name" for message types which can then be referred to by buffer and join operations for cases where the target message type can't be reasonably inferred.

Ultimately we decided to pass on this idea because stable names are tricky to get right. We managed to implement the majority of buffer and join use cases without it.

However I've found situations where there's no way for a buffer operation to infer its target. If all the nodes using a buffer are using buffer access to put values in and out then the buffer itself cannot infer what its underlying data type is. In this situation the only way to make it work is to specify the message name for the buffer.

Even trying to specify a target accessing node wouldn't work because a node could access many different buffers simultaneously. In theory there might be convoluted ways to infer it with backward propagation, but this will probably be difficult to implement and confusing for users.

After #48 is finished we should revisit this feature in a follow-up PR.

Start by reading the linked PR 51 discussion and issue #48, then trace the buffer and join operations. Define how a buffer can accept an explicit message type when inference is impossible, while preserving existing inference behavior. Done means buffer access works for the described case and the behavior is covered by the project's relevant tests.