onnxruntime

b3d7009d - [CUDA EP] Add kernel registrations for Pad and ConvTranspose opset 19-23

[CUDA EP] Add kernel registrations for Pad and ConvTranspose opset 19-23 Fixes #26393 This PR adds CUDA kernel registrations for newer ONNX opset versions: **Operators Updated:** - Pad: Extended support from opset 18 to opset 19-23 (float, double, MLFloat16, bool) - ConvTranspose: Extended support from opset 11 to opset 11-23 (float, double, MLFloat16) **Changes Made:** 1. cuda_execution_provider.cc: Added forward declarations and BuildKernelCreateInfo for new opsets 2. cuda_nhwc_kernels.cc: Updated NHWC domain registrations for ConvTranspose 11-23 3. nn/conv_transpose.cc: Changed opset 11 from TYPED to VERSIONED_TYPED (11-23) 4. tensor/pad.cc: Extended opset 18 to VERSIONED (18-23) **Testing:** - All 17 ONNX backend tests pass with CUDAExecutionProvider - Pad opset 24: 6 tests passed - ConvTranspose opset 22: 11 tests passed - No unresolved symbols - CUDA EP loads successfully The kernel implementations already existed and supported these opset versions; only the forward declarations and registry entries were missing.