add madd_no_fma_helper

Modifies the general case of gebp_traits so that the tmp variable of madd in the no_fma case is only used if its type, RhsPacketType is compatible with (able to store the result of) AccPacketType, otherwise a fresh AccPacketType is used instead.

Why do this at all? AccPacketType is, approximately, the result of LhsPacketType * RhsPacketType, and the existing code attempts to store this type inside tmp. However, tmp has the type RhsPacketType, so unless AccPacketType can be converted and stored into RhsPacketType, a compilation error results. Alternative methods, such as altering the type of the argument of tmp to AccPacketType necessitate spillover changes elsewhere, since both B# and T0 are provided as arguments for tmp and the function signatures become ambiguous. It is also painful for other projects to specialize the (medium size, well tied in) gebp_traits template. The change here is minimal and should create no overhead for the cases that already worked under the old code. In the new use cases it enables, it could be more efficient by reusing the temp variable via (thread-local) static, or not assigning from b, etc. However, I wanted to keep the changes as simple as possible.