fix(numpy): numpy annotation processing

[judahrand]

What does this PR address?

Currently, it seems to me that numpy annotations are broken. For example, neither of the correct types npt.NDArray[np.float32] or np.ndarray[Any, np.dtype[np.float32]], are processed and instead result in errors.

These changes fix this and add to the unit tests to catch regressions. Additionally, I've made sure that if the type hint for a Numpy array doesn't match the Annotated DType then we throw an error.

I'm really surprised how little unit testing there is in this area - am I missing something?

Fixes #(issue)

Before submitting:

• Does the Pull Request follow Conventional Commits specification naming? Here are GitHub's
  guide on how to create a pull request.
• Does the code follow BentoML's code style, pre-commit run -a script has passed (instructions)?
• Did you read through contribution guidelines and follow development guidelines?
• Did your changes require updates to the documentation? Have you updated
  those accordingly? Here are documentation guidelines and tips on writting docs.
• Did you write tests to cover your changes?

[judahrand]

Hmmm... looks like Python 3.8 isn't happy. Will look into it.

[judahrand]

Hmmm... looks like Python 3.8 isn't happy. Will look into it.

I think I got to the bottom of this. There is some weirdness in how get_origin and get_args work on Python 3.8 vs 3.11. It seems that if we try foo.__origin__ and foo.__args__ first we should get the right result.

[aarnphm]

LGTM.