This ticket captures a few issues that were seen in math intrinsic lowering.

1. llvm intrinsic lowering table contain entries for calls to the math library for atan, cosh, sinh.
2. By default math intrinsics are outlined, this can impact vectorisation since the outlined function name is not known to the llvm vectoriser but the llvm intrinsic name is known.
3. There is no support for intrinsics with complex inputs in llvm lowering.
4. There is no support for kind=10,16 in generic/pgmath lowering.