onnxruntime

1953d04a - Make element-wise fp16 tests skip explicitly on CPU-only builds

Make element-wise fp16 tests skip explicitly on CPU-only builds TestBinaryFloat16/TestUnaryFloat16 only run when a CoreML or CUDA EP is compiled in; otherwise they silently did nothing while the enclosing float32 test still reported success, hiding that the fp16 path was never exercised. Because these helpers were called inside the combined float32 tests, a helper-level GTEST_SKIP would have hidden the float32 coverage too. Instead, extract the fp16 calls into dedicated <Op>_float16 tests guarded by '#if defined(USE_CUDA) || defined(USE_COREML) ... #else GTEST_SKIP()', matching the existing Pow_float16_float16 pattern. On CPU-only builds the float32 tests still pass and the fp16 tests now report SKIPPED.