Fix for multiplying an integer unit type by a float

[JonSalmon-CC]

In the following situation, the library does not behave as expected:

constexpr SI::gram_t<uint32_t> mass = 1000_g;

SI::gram_t<uint32_t> half_mass = mass * 0.5f;
cout << half_mass.value(); // Prints out 0!

Without this fix, the library silently down-cast the float to an integer (i.e. zero) before the multiplication. With this fix it won't, although some compilers may warn about the int * float multiplication.