Bounds checks for IntegerArray

[TomAugspurger]

On the current PR there are a couple "issues" with handling out of bounds values.

This follows NumPy's behavior, but is maybe not what the user wants

In [18]: pd.core.arrays.IntegerArray([-1, 2, 3], dtype=pd.core.arrays.integer.UInt16Dtype)
Out[18]: IntegerArray([65535, 2, 3], dtype='UInt16')

In [19]: x = pd.core.arrays.IntegerArray([-1, 2, 3], dtype=pd.core.arrays.integer.UInt16Dtype)

In [20]: x[0] = -1

In [21]: x
Out[21]: IntegerArray([65535, 2, 3], dtype='UInt16')

These could perhaps raise a ValueError instead.

xref #21160

[mroeschke]

Looks like this correctly raises a TypeError now. Could use a test

In [7]: pd.array([-1, 2, 3], dtype='UInt16')
TypeError: cannot safely cast non-equivalent int64 to uint16

[TAJD]

take