Rich future tools use double in vectorised ray tracing

Switches from the use of float to double SIMD types in the vectorised ray tracing. Done to mitgrate differences seen in results between the difference SIMD implementations (e.g. AVX versus SSE4 versus AVX2+FMA). Results in a small regression in the speed of the ray tracing algorithm, but given the algorithm is not yet fully vectorised, it actually fairly modest (O(%) on the full RICH sequence) so I think, until futher work is done to improve things worth the cost.