Add datetime testing skeleton

[tacaswell]

This is prep work for the 2023 GHC OSD event.

Although we nominally support units through all of our plotting APIs, this is not actually tested. As the most common "unitful" data is likely datetime, this is using datetime as the vehicle to exhaustively test which of the methods correctly handle units.

The ask of the volunteers will be to pick a method and write/find an example that passes datetime64 or timedelta64 arrays to as many of the inputs as make sense. Then

1. if this works, open a PR adding the implementation as a test to this file + the output of the example when run as an image
2. if this fails decide if:
   1. it should not work (for example ax.pie makes no sense with dates as the input so I pre-emptively removed it...although maybe with timedelta?) open a PR removing that stub test
   2. if it should work open an issue with the failing example

Unfortunately I did not get the test image overhaul done in time so I want to avoid adding test images for now, but smoke tests are still an improvement and once that lands it should be possible to "just" add a decorator to the tests to generate the images in the future.

[story645]

ax.pie makes no sense with dates as the input so I pre-emptively removed it...although

timedeltas though totally - thinking particularly of the American time use survey

[tacaswell]

put pie back

[tacaswell]

If we merge this I will write up a tracking issue with checklist of the methods + links etc for tomorrow.

[QuLogic]

Should you also test ax.plot(range(1, N), x) and/or ax.plot(x, x)? It might be good to give an example of what to do when multiple arguments accept units.

[tacaswell]

which....looks not great

[oscargus]

I think we can skip psd etc. I cannot really see a way to use dates there.

[tacaswell]

That looks better.