BUG: inconsistency between str.center and np.char.center #25659

Note that the same issue happens with ljust, rjust, and zfill or any other operation implemented using _vec_string.

The python docs for str.center say:

The original string is returned if width is less than or equal to len(s).

And the docs for np.char.center say:

Calls :meth:`str.center` element-wise.

but observe:

In [1]: np.char.center(["this is a string that's longer than the width", "a"], 10)
Out[1]: array(['this is a ', '    a     '], dtype='<U10')

In [2]: "this is a string that's longer than the width".center(10)
Out[2]: "this is a string that's longer than the width"

Now I can see why the _vec_string implementations (and probably future ufunc implementations from @lysnikolaou) are written this way. I'm perfectly happy to keep this difference compared with str.center, but we should intentionally choose that and document it in the affected functions. I'm happy to implement whichever behavior we decide on in stringdtype.