process: make sure type annoations pass with mypy

[plamut]

Towards #536.

This is a draft PR. The new noxfile check, mypy, passes locally, but there are still a few things to consider:

• The generated code is omitted from type checks, because mypy reports several errors for it
  (will be done separately, since it's currently blocked by Generated code for Pub/Sub does not pass type checks with mypy gapic-generator-python#1092 )
• The PR adds py.typed marker file, but we might want to postpone this addition until the generated code is error-free. Marker file removed for the time being.
• Some type aliases such as _TimeoutType are defined in multiple places. We should probably extract the definition to a single central place within the library.

PR checklist:

• Make sure to open an issue as a bug/issue before writing your code! That way we can discuss the change, evaluate designs, and agree on the general idea
• Ensure the tests and linter pass
• Code coverage does not decrease (if any source code was changed)
• Appropriate docs were updated (if necessary)

[tseaver]

Overall, I'd like to see us using only one typing checker, and I think 'mytype' is the consensus favorite.

[tseaver]

I don't understand the type: ignore here, and I think this may be a misuse of Type[].

[plamut]

The ignore here is needed, because PubsubMessage is dynamically injected into google/cloud/pubsub_v1/types.py and both type checkers think it does not exist. The alias was created to not repeat the same ignore comment in every line where message is a method parameter.

Using a plain types.PubsubMessage in the alias results in mypy complaining that variable MessageType is not valid as a type. Using Type[...] works around that.

(alternatives welcome, especially if they are canonical)

[plamut]

A few trivial errors, will fix the checks tomorrow.

[plamut]

@tseaver The required changes turned out to be more than just a few trivial lines, thus please take another look when you manage.

It's mostly just getting rid of cast() for Optional types, and one extra case in Dispatcher had to be covered after splitting batch_commands into separate lists (coverage complaint).