Fix compile-time concatenation with negative right operand

[vlstill]

Fixes #5231. The problem is most likely in the binary representation of the big_int for negative numbers. The solution is to always convert the right operand to a unsigned value with the same representation as does the signed value have.

[jafingerhut]

The test P4 program and its expected output looks correct to me.

[ChrisDodd]

There's no need to create a new IR::Constant here that you're going to throw away -- just take the big_int value and fix it if it is negative:

big_int rvalue = right->value;
if (rvalue < 0)
    rvalue &= Util::mask(rt->size);

and then use rvalue later instead of right->value

[vlstill]

I do not like the idea of spreading a conversion function from 2-complement signed to 2-complement unsigned over multiple places in the codebase. It is already in the Constant::handleOverflow (invoked from the ctor) and it is not really possible to extract it from there because of the warnings which use the constant value in the logs.

I've changed the code to avoid the unnecessary dynamic allocation at least.

[ChrisDodd]

IMO this is more obscure than just doing the mask with the big_int value. boost::mulitprecision specifies that bitwise operations on negative value happen as-if the values are 2s complement (regardless of what the actual underlying representation is), so that does what we want for concatenation according to the P4 spec.