onnxruntime

306cd988 - Add CreateTensor with byte offset support for WebGPU sub-buffer views

Add CreateTensor with byte offset support for WebGPU sub-buffer views - Add CreateTensorWithDataAsOrtValueWithByteOffset to OrtApi (v26, offset 418) + WebGPU branch: stores raw WGPUBuffer handle and offset via CreateTensorImplWithByteOffset + CPU/CUDA branch: advances pointer directly before delegating to CreateTensorImpl - Add typed (element offset) and void* (byte offset) C++ wrapper overloads in onnxruntime_cxx_api.h / onnxruntime_cxx_inline.h - Update WebGPU DataTransfer (data_transfer.cc) to recover base WGPUBuffer from DataRaw() - ByteOffset() for all three copy directions (GPU->GPU, CPU->GPU, GPU->CPU) - Update LaunchComputePipeline to accept bind_buffers_byte_offsets for per-binding WGPUBindGroupEntry offsets in webgpu_context.h / webgpu_context.cc - Remove indirect_buffer_offset from CapturedCommandInfo: indirect buffers are always allocated internally by the WebGPU EP and never have a non-zero ByteOffset() - Add tests in test_data_copy.cc covering typed/void* offset overloads and CopyTensors