Unify input/output types

[mrwyattii]

Several changes in this PR, with the main goal to make the inputs/outputs of pipeline and client.generate match. That is, each can take str|List[str] as input and will return List[Response] where Response is mii.batching.data_classes.Response.

In addition, I've modified the entire chain of communication between the MII user-facing APIs to the inference server to better represent the inputs and outputs. This includes removing many protobuf message types that are using in MII-Legacy, but not the latest MII.

Full summary of changes:

• Move Response, Request, etc. classes to new data_classes.py file
• Made the output of client.generate match that of pipeline
• Removed old GRPC/proto code that is unused
• Changed how we represent response messages to better capture generation details

Note: This looks like a lot of change, but it's mostly moving things around the recompiling the proto stuff

TODO:

• Test pipeline
• Performance testing
• Docs
• Update DS and DSExamples (No changes for DS, update MII benchmark DeepSpeedExamples#812)

[tohtana]

This achievement is great, thank you!
Since we don't have response.response any more in the client code, we need to make sure we update the benchmark scripts in DeepSpeedExamples after this PR is merged or the package on PyPI is updated