Refactor StreamInferHandler for better readability

[rmccorm4]

This PR does two things:

1. Refactor the handling of every step in StreamInferHandler::Process from one large ~400 line function into helper functions for each state.
2. Replace the "double negative" logic of if not not_finished into just if finished in Common/InferHandler/StreamInferHandler Process() functions. I found the double negative logic confusing that the function did:

bool Process(...) {
  ...
  finished = true; 
  // (not finished)
  return !finished;
}

and then the caller checks the negation of that:

// if (not (not finished) )
if (  !Process(...) )

So instead if's more direct to be:

bool Process(...) {
  finished = true;
  return finished;
}

if ( Process(...) {
  ...
}

[tanmayv25]

Can we pull the step transition from within these helpers to this Process function? It would help in giving a complete picture of the state machine being implemented here.
If not, then may be add a comment describing to which state the state can transition to in these helper functions.

[rmccorm4]

This would be nice, but there's a lot of branching in each function where if (this) state = state1; else state = state2; It would be non-trivial effort to extract this out I think.

I added some descriptive comments for now.

[rmccorm4]

@tanmayv25 I don't understand this part. Do you know why we try to Write the next response (not the current' state's response) here in the WRITTEN state for a non-decoupled request?

For the non-decoupled case, isn't every response's write handled when calling the response callback or error handling for a given request?

[rmccorm4]

Also @tanmayv25 , I don't really understand the point of the COMPLETE state/step. Why don't we directly transition to FINISH state anywhere we would transition to COMPLETE instead? Seems like an unnecessary intermediate step.

[rmccorm4]

Closing due to lots of changes around request cancellation. May come back to this later.