fix!: use pandas.NaT for missing values in dbdate and dbtime dtypes

[tswast]

This makes them consistent with other date/time dtypes, as well as internally
consistent with the advertised dtype.na_value.

BREAKING-CHANGE: dbdate and dbtime dtypes return NaT instead of None for missing values

Release-As: 0.4.0

Thank you for opening a Pull Request! Before submitting your PR, there are a few things you can do to make sure it goes smoothly:

• 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)

Fixes #66
🦕

[shollyman]

Unclear how/why this gets used in practice. Is this an attempt at exposing typed nulls in the pandas space?

[tswast]

Numpy/Pandas have a special case for date/time arrays. The underlying c-buffer is of integers, rather than floating point values, so NaN can't be stored. Instead they pick a date/time that's unlikely to be used and add lots of checks throughout the code so that it behaves mostly like a NaN would.

https://github.com/numpy/numpy/blob/72a27ad2798575fa5dce9046815597b3a63109a2/numpy/core/include/numpy/ndarraytypes.h#L239

Some new Pandas types also have a pandas.NA value, but this is achieved via a bit mask for null/not-null. Since we're backing our arrays with a numpy datetime64 array, NaT is what we need to store missing values as. The inconsistency between what we were storing and what we were exposing to users was causing some edge cases to break, such as trying to see if an array contains a missing value.