[HLSL] implement `sqrt` intrinsic

[farzonl]

The current sqrt intrinsic is incomplete in some cases and not correct in others.

For example there is no double support. Doubles need to cast down to floats:
https://godbolt.org/z/Pc73srEr7
Yet we added it here: https://github.com/llvm/llvm-project/blob/main/clang/lib/Headers/hlsl/hlsl_intrinsics.h#L1421

Second we should be using __builtin_elementwise_sqrt and not multiple different builtins especially since all sqrt builtins map to the same Intrinsic::sqrt
https://github.com/llvm/llvm-project/blob/main/clang/lib/CodeGen/CGBuiltin.cpp#L2908

Using one builtin makes it easier to apply type check rules in sema.

Finally lowering support needs to be added. Sqrt lowering is very simple and has a one to one mapping between

Intrinsic::sqrt and dx.op.unary 24