Enable multi-device test for GPU * Add build pipeline for TensorRT multi-GPU test * Add code to disable fp16 test if hardware architecture not supported * Add option to set the device id in onnx_test_runner for model tests
