Decide if to handle -infinity + infinity addition correctly.

[libbooze]

I believe IEEE says that -inf + inf should produce NaN, but dfp produces garbage, e.g.

    auto inf = __bid64_from_string("INF", 0, &flags);
    std::cout << "inf: " << str(inf) << "  " << str(-inf + inf) <<  std::endl;

prints

inf: +Inf  +0E-398

for Decimal

template<typename T>
void nan_inf_add()
{
    using nl = std::numeric_limits<T>;
    auto inf = nl::infinity();
    std::println("{}, {}",inf,  -inf+inf);
}

prints

inf, -inf

while float and double print

inf, -nan

I am not sure how valuable is to check for this since addition is performance critical operation, but if this was not intentional then maybe Decimal should produce Nan(or -Nan).
Also I did not find the IEEE authoritative reference for inf addition, so if you have a document that specifies this result for FP numbers or decimal FP in particular feel free to correct my assumptions.

[mborland]

IEEE 754-2008 Section 7.2.d says a quiet NaN should be returned from: "Addition or subtraction or fusedMultiplyAdd: magnitude subtraction of infinities, such as: addition(+inf, -inf)" so you are correct. I'll have to relook the other edge cases from 7.2 as well.