onnxruntime

77637752 - Remove CopyTensorsEx; keep CreateTensor(+offset) approach for sub-buffer views

Remove CopyTensorsEx; keep CreateTensor(+offset) approach for sub-buffer views CopyTensorsEx is superseded by CreateTensorWithDataAsOrtValueWithByteOffset: callers now create an offset tensor view once and pass it to regular CopyTensors. Changes: - Remove OrtApi::CopyTensorsEx entry and vtable slot (offset 417 freed); CreateTensorWithDataAsOrtValueWithByteOffset moves to offset 417 - Remove CopyTensorsEx forward decl from ort_apis.h and both implementations (full build and minimal-build stub) from onnxruntime_c_api.cc - Remove source_offsets/destination_offsets/sizes params from OrtDataTransferImpl::CopyTensors in onnxruntime_ep_c_api.h (EP C API) - Revert plugin_data_transfer.cc to not propagate offset arrays - Revert framework IDataTransfer: remove CopyTensor/CopyTensorAsync overloads with offset params; remove source_offset/destination_offset/size from SrcDstPair; simplify CopyTensors loop back to original; remove CPUDataTransfer offset overload - Revert WebGPU DataTransfer: remove CopyTensor(+offsets) overload; fold ByteOffset recovery into single CopyTensor(src,dst) which uses tensor's own ByteOffset() to recover the base WGPUBuffer and pass correct offsets to BufferManager::MemCpy/Upload/Download (buffer_manager offsets kept) - Revert webgpu_provider_factory.cc CopyTensorsImpl to simpler signature and call CopyTensor(src,dst) directly - Revert example EP test files (ep_data_transfer.h/.cc x2, utils.h)