[BUG] Including a NaN (float) value should force Index to cast to float64.

[bdice]

Describe the bug
Constructing an integer index with a NaN value produces an index of ints with a null value, rather than casting to floats.

Steps/Code to reproduce bug
This doctest (#9815) is failing:

cudf/python/cudf/cudf/core/_base_index.py

Lines 832 to 834 in 582cc6e

	>>> idx = cudf.Index([1, 2, 3, 4, np.nan])
	>>> idx.is_floating()
	True

The index is actually getting this value:

>>> cudf.Index([1, 2, 3, 4, np.nan])
Int64Index([1, 2, 3, 4, <NA>], dtype='int64')

Expected behavior
NaN is a float value, and thus should force the column to cast to floats. This would match the expectation of other float values like 3.5 (instead of NaN). This is what pandas does:

>>> pd.Index([1, 2, 3, 4, np.nan])
Float64Index([1.0, 2.0, 3.0, 4.0, nan], dtype='float64')