Horizontal reductions with scalar initializer

[edanor]

All reduction operations should also have a version with a scalar initializer. These operations are necessary to improve the programmability. For example to perform a hoirzontal reduction with a scalar value, the user has to write:

SIMDVec<float, 4> vec(...);
float init_value = ...;
float max_value = std::max(vec.hmax(), init_value);

With the scalar initializers one could just write:

float max_value = vec.hmax(init_value);

I've already extended the interface with HMAXS/MHMAXS, but the changes have to be propagated to all plugins and other reduction operations.