[SYCL][CUDA][libclc] Implement nextafter for sycl::half in generic/.

[hdelan]

sycl::nextafter(half,half) was defaulting to sycl::nextafter(float,float) which does not return the next half.

Software implementation written in libclc/generic and #included into ptx-nvidiacl.

[bader]

@alexbatashev, pre-commit for this PR failed with

sycl/tools/pi-trace/pi_trace.cpp
sycl/tools/pi-trace/pi_trace.cpp:32:1: error: ‘XPTI_CALLBACK_API’ does not name a type; did you mean ‘CL_CALLBACK’?
 XPTI_CALLBACK_API void tpCallback(uint16_t trace_type,
 ^~~~~~~~~~~~~~~~~
 CL_CALLBACK
sycl/tools/pi-trace/pi_trace.cpp:40:1: error: ‘XPTI_CALLBACK_API’ does not name a type; did you mean ‘CL_CALLBACK’?
 XPTI_CALLBACK_API void xptiTraceInit(unsigned int /*major_version*/,
 ^~~~~~~~~~~~~~~~~
 CL_CALLBACK
sycl/tools/pi-trace/pi_trace.cpp:66:1: error: ‘XPTI_CALLBACK_API’ does not name a type; did you mean ‘CL_CALLBACK’?
 XPTI_CALLBACK_API void xptiTraceFinish(const char * /*stream_name*/) {
 ^~~~~~~~~~~~~~~~~
 CL_CALLBACK
sycl/tools/pi-trace/pi_trace.cpp:70:1: error: ‘XPTI_CALLBACK_API’ does not name a type; did you mean ‘CL_CALLBACK’?
 XPTI_CALLBACK_API void tpCallback(uint16_t TraceType,
 ^~~~~~~~~~~~~~~~~
 CL_CALLBACK
sycl/tools/pi-trace/pi_trace.cpp:25:16: error: ‘GStreamID’ defined but not used [-Werror=unused-variable]
 static uint8_t GStreamID = 0;
                ^~~~~~~~~
cc1plus: all warnings being treated as errors

I have no idea why this issue appears for libclc changes. Could you take a look, please?