Now that `sincos` is a supported intrinsic in the LLVM dialect (#160561) we are able to add the corresponding operation in the math dialect and add conversion patterns for LLVM and NVVM. We have several benchmarks that use sine and cosine in hot-loops, and saving some calculations by performing them together can benefit performance. We would like to have a way to represent sincos in the math dialect.