Add __round__ for .NET integer types

[slozier]

Adds a __round__ implementation for the .NET integer types (e.g. System.Int16, System.UInt16, etc.) with the following semantics:

• Calling __round__ without an ndigits argument always returns a Python int.
• Calling with an ndigits argument will attempt to preserve the type. In the case where an overflow would occur I chose widen to a Python int instead of overflowing which is consistent with the other operators we implement.

Note that numpy behaves similarly except that it overflows instead of widening. I think I would have preferred this behaviour but chose to be consistent with the other operators.

Resolves #1977