Use __index__ if needed to pack integer values in struct

[BCSharp]

Resolves #1381.

[slozier]

Looks good to me. Interesting that CPython has two different error messages for some of these.

Random thought, I wonder if trying to cast the BigInteger and catching the failure would be more performant than checking the range.

[BCSharp]

Intuitively, a direct cast should be faster for cases without overflow, but I was curious by how much, so I did some measurements.

The results show that going through AstInt32 adds ca. 550% to the execution of the cast (i.e., the whole operation is 6.5 times slower). About half of this overhead comes from the two BigInteger comparisons, and the other half from the static method call (and whatever else that method does).

That said, those things are extremely fast (and hard to measure accurately). Converting by calling AstInt32 on my machine is less than 8 nanoseconds. So in the end, it is a matter of judgment if shaving off a few nanoseconds is worth increased code verbosity.