improve ISO 8601 parsing for temporal filter

[itcarroll]

Fixes #330.

As suggested by @jhkennedy, earthaccess will now pass a datetime object (or None) to cmr, essentially replacing the basic string parsing that cmr does (using datetime.strptime) with flexible and timezone aware parsing by dateutil.parser.

The change created the opportunity to create a new suite of tests for possible user inputs, that includes checking exceptions.

---

📚 Documentation preview 📚: https://earthaccess--486.org.readthedocs.build/en/486/

[itcarroll]

Believe the pre 3.10 checks are failing due to None | dt.datetime as a type hint for the return, which must be newer syntax.

[jhkennedy]

Believe the pre 3.10 checks are failing due to None | dt.datetime as a type hint for the return, which must be newer syntax.

Yes, using | instead of typing.Union is new as of Python 3.10. In general, you can write | like:

from typing import Union

def foo(bar: Union[int, float]):

For None | thing, specifically, there's a special Optional type:

from typing import Optional

def fizz(buzz: Optional[thing]):   # buzz can be None or thing

But that's an easy circle back once we're happy with the implementation

[itcarroll]

Test error is ModuleNotFoundError: No module named 'numpy'. How could the test environment be missing numpy?

[jhkennedy]

What I don't love about the simplification here is the call to parse when it may not have been necessary (if we already had a datetime).

I think that roundtrip significantly simplifies the logic in the function; your latest version has nested branches/leafs which I find makes it harder to follow the logic all the way through without spending some time in the function.

That said, we've got really good tests now, so I'm fine with whatever implementation you prefer that passes the tests.

[itcarroll]

With nasa/python_cmr#31 merged, I believe we can close this as a won't fix in favor of a new PR that passes user's temporal input directly to python_cmr. Sound good?