Validate that MultiIndex labels are Int64 type when not called internally.

[jtratner]

Right now you could create MI with weird types and it doesn't error out. More difficult to do effectively when MI is an ndarray subclass (because there are so many different ways to instantiate a MultiIndex). Will fix in Index refactor (just want to remember to do so).

[toobaz]

@jtratner does #19074 change anything?

[mroeschke]

It appears that codes are dtyped resized internally

In [69]: pd.MultiIndex(levels=mi.levels, codes=np.array([[0., 0., 1., 1.], [1., 0., 1., 0.]], dtype=np.in
    ...: t32)).codes[0]
Out[69]: array([0, 0, 1, 1], dtype=int8)

So this is probably not an issue anymore. Closing