[libclc] Add generic native half implementation of __clc_normalize

[wenju-he]

This is ported from https://github.com/intel/llvm/blob/sycl/libclc/libspirv/lib/generic/geometric/normalize.cl and can pass a closed-source OpenCL CTS
"test_geometrics geom_normalize --half CL_DEVICE_TYPE_GPU" on intel GPU.

llvm-diff amdgcn--amdhsa.bc shows fpext/fptrunc insts are now removed from normalize function.

[wenju-he]

llvm-diff amdgcn--amdhsa.bc.new amdgcn--amdhsa.bc.old

in function _Z9normalizeDh:
  in block %1 / %1:
    >   %2 = fpext half %0 to float
        %2 = fcmp one half %0, 0xH0000
    >   %4 = select i1 %3, float 1.000000e+00, float 0.000000e+00
    >   %5 = tail call noundef float @llvm.copysign.f32(float %4, float %2)
    >   %6 = fptrunc float %5 to half
    >   ret half %6
    <   %3 = select i1 %2, half 0xH3C00, half 0xH0000
    <   %4 = tail call noundef half @llvm.copysign.f16(half %3, half %0)
    <   ret half %4

in function _Z9normalizeDv2_Dh:
  in block %1 / %1:
    >   %2 = fpext <2 x half> %0 to <2 x float>

in function _Z9normalizeDv3_Dh:
  in block %1 / %1:
    >   %2 = fpext <3 x half> %0 to <3 x float>

in function _Z9normalizeDv4_Dh:
  in block %1 / %1:
    >   %2 = fpext <4 x half> %0 to <4 x float>