This is what rust-geom uses, and it seems to make more sense. I originally used _t because it referred to the base generic type, but I think _s makes more sense.
These operations could potentially mess up the orthoganality of the rotation matrix. We could have made these fail if an the resulting matrices were not orthogonal, but it is preferable to enforce this at compile time rather than run time. More operations should be added in the future with this restriction in mind.
Having these two modules separate made more sense when the library was divided into lmath and cgmath. Now that the two have been combined, it makes more sense to bundle these types in one module (I think).