Categorical handles imaginary values incorrectly

[TomAugspurger]

In [3]: import pandas as pd

In [4]: pd.Categorical([1, 2 + 2j])
Out[4]:
[1.0, 2.0]
Categories (2, float64): [1.0, 2.0]

In [6]: pd.Categorical([1, 2, 2 + 2j])
Out[6]:
[1.0, 2.0, 2.0]
Categories (2, float64): [1.0, 2.0]  # Missing a `2 + 2j` category

I think it comes down to factorize

In [9]: pd.factorize([1, 2, 2 + 1j])
Out[9]: (array([0, 1, 1]), array([ 1.,  2.]))

Clearly not much demand for this, but figured I'd park an issue anyway. Discovered this in #16015, which will not be fixing it.

[jreback]

we convert complex to float (always have)

[TomAugspurger]

we convert complex to float (always have)

Yeah, the float part is fine, it's the dropping the imaginary part (Out[6] should have three categories)

[jbrockmendel]

I think this now works since we support Index[complex]. Could use test(s).