Numeric Index constructors should validate `dtype` parameter

[jschendel]

The Int64Index, UInt64Index, and Float64Index constructors do not validate the dtype parameter and instead ignore it when it conflicts with the default dtype:

In [1]: import pandas as pd; pd.__version__
Out[1]: '0.26.0.dev0+842.g78c3c1640'

In [2]: pd.Int64Index([1, 2, 3], dtype="float64")
Out[2]: Int64Index([1, 2, 3], dtype='int64')

In [3]: pd.UInt64Index([1, 2, 3], dtype="categorical")
Out[3]: UInt64Index([1, 2, 3], dtype='uint64')

In [4]: pd.Float64Index([1, 2, 3], dtype="datetime64")
Out[4]: Float64Index([1.0, 2.0, 3.0], dtype='float64')

Other index types do not allow non-standard dtypes:

In [5]: pd.CategoricalIndex(list('abc'), dtype='timedelta64')
---------------------------------------------------------------------------
ValueError: Unknown dtype 'timedelta64'

In [6]: pd.DatetimeIndex(['2018-01-01'], dtype='int64')
---------------------------------------------------------------------------
ValueError: Unexpected value for 'dtype': 'int64'. Must be 'datetime64[ns]' or DatetimeTZDtype'.

In [7]: pd.IntervalIndex([pd.Interval(0, 1)], dtype=object)
---------------------------------------------------------------------------
TypeError: dtype must be an IntervalDtype, got object

In [8]: pd.RangeIndex(0, 3, dtype='float64')
---------------------------------------------------------------------------
TypeError: Invalid to pass a non-int64 dtype to RangeIndex

Expected Output

I'd expect In [2]/In [3]/In [4] to raise a ValueError.

[MarcoGorelli]

Sure, I could work on this if you like (though if any new contributors following along would like to take it, please do go ahead)