Regular indexing behaves like jagged index for RegularArray (v2)

[agoose77]

Version of Awkward Array

c2ddda2

Description and code to reproduce

>>> array = ak.from_numpy(np.arange(4 * 4).reshape(4, 4))
>>> array[array > 4].ndim
2
>>> (array > 4).layout.purelist_isregular
True
>>> array[ak.to_numpy(array > 4)].ndim
1

Cause

The NumPy-like indexing checks for NumpyArray, not RegularArray. We should use something like maybe_toNumpyArray beforehand to fulfil this.

[agoose77]

What should array[EmptyArray()] return? Currently it is returning a NumpyArray with int64. This seems wrong to me, but there is no clear idea of what would be right. Maybe this should be an error?

[jpivarski]

array[[]]

should interpret the typeless list as though it were an array of integers, and the return value would be an empty array with the same type as array.

If the slicer has a type, then we use that: boolean, integers, strings.