Most modern processors support instructions for taking the minimum or maximum of two floating point values, avoiding the branching that would occur in expressions such as `if x >= y then x else y`. This ticket requests the addition of new primops for taking the minimum of two floating-point values, such as `minDouble# :: Double# -> Double# -> Double#`. These should be compiled to native instructions when possible, e.g. `minss`, `maxss`, `minsd`, `maxsd` on X86 with SSE2.