TST: Remove arbitrary pytest markers and replace ccd_data fixture

As @bsipocz pointed out, pytest emits the following warning in nddata tests (turned into error in this exercise), which we currently ignore:

__ ERROR collecting astropy/nddata/tests/test_ccddata.py __
astropy/nddata/tests/test_ccddata.py:98: in <module>
    @pytest.mark.data_size(10)
.../_pytest/mark/structures.py:337: in __getattr__
    PytestUnknownMarkWarning,
E   _pytest.warning_types.PytestUnknownMarkWarning: Unknown pytest.mark.data_size
- is this a typo?  You can register custom marks to avoid this warning
- for details, see https://docs.pytest.org/en/latest/mark.html

This is because of

https://github.com/astropy/astropy/blob/a22c724006036c1ddf5e68aa896ae90bf7e0b6a0/astropy/nddata/tests/test_ccddata.py#L38-L43

https://github.com/astropy/astropy/blob/a22c724006036c1ddf5e68aa896ae90bf7e0b6a0/astropy/nddata/tests/test_ccddata.py#L46-L47

https://github.com/astropy/astropy/blob/a22c724006036c1ddf5e68aa896ae90bf7e0b6a0/astropy/nddata/tests/test_ccddata.py#L64

https://github.com/astropy/astropy/blob/a22c724006036c1ddf5e68aa896ae90bf7e0b6a0/astropy/nddata/tests/test_ccddata.py#L98-L99

I think the tests would still work if we remove ccd_data as a fixture, but rather re-implements it as a helper function that takes all those custom markers as regular arguments. Better yet, we can memoize that function so that requesting array of same size will return a cached copy instead of creating a new one from scratch (more memory usage but better performance). Given that ccd_data generates random number and there is no reason why randomness has to be that random between different unit tests, memoization is probably okay?

What do you think, @mwcraig , @crawfordsm , or @MSeifert04 ?

[MSeifert04]

Okay, seems like we all agree. I could have a look into this if none of you is already working on that.

[mwcraig]

go for it!