clean up usage of _is_na, _is_true and _is_false

[grst]

Description of feature

Since #356, we have additional guarantees about the data types in .obsm["airr"] and there is no danger that AnnData changes the data type of a pandas column anymore during save/load. Therefore, we can get rid of most instances of _is_na, _is_true, and _is_false checks. The only remaining instances should be in the IO module sanitizing input before it is stored in AnnData.

Cleaning this up should improve performance, and make the code more legible.

When doing so, tests should be added that everything still works as expected when creating AirrCell objects from data that does contain string representations of NaN/True/False/...