[webgpu] Optimize generic 4D Transpose using OIHW2OHWI Program

[daijh]

Description

This PR migrates the OIHW2OHWI Program from Im2ColMatMul to the Transpose operator. By centralizing this logic, we leverage the specialized shader to optimize generic 4D transpositions (specifically the {0, 2, 3, 1} permutation pattern) while reducing code duplication.

While this shader is capable of supporting 2D/3D transpositions, those optimizations are reserved for follow-up PRs.

Motivation and Context

See above.

[daijh]

@guschmue @fs-eire @qjia7 PTAL

[Copilot]

Pull request overview

This PR optimizes generic 4D transpose operations by migrating the specialized OIHW2OHWIProgram shader from the Im2ColMatMul operator to the Transpose operator. The migration enables reuse of this optimized shader for any 4D tensor transpose with the {0, 2, 3, 1} permutation pattern, while also fixing a calculation bug in the process.

• Moves OIHW2OHWIProgram class and implementation from im2col_matmul to transpose
• Relocates the WGSL shader template from nn/ to tensor/ directory
• Fixes a bug where H_W_tiles was calculated using kernel_height * kernel_height instead of kernel_height * kernel_width

Reviewed changes

Copilot reviewed 4 out of 5 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
onnxruntime/core/providers/webgpu/tensor/transpose.h	Adds OIHW2OHWIProgram class declaration with uniform variable definitions
onnxruntime/core/providers/webgpu/tensor/transpose.cc	Implements OIHW2OHWIProgram shader generation and integrates the optimization into DoTranspose with proper threshold checks; fixes bug in H_W_tiles calculation
onnxruntime/core/providers/webgpu/tensor/oihw_to_ohwi.wgsl.template	Adds the WGSL shader template for the OIHW to OHWI transpose operation with proper bounds checking and workgroup synchronization
onnxruntime/core/providers/webgpu/nn/im2col_matmul.h	Removes OIHW2OHWIProgram class declaration as it's moved to transpose
onnxruntime/core/providers/webgpu/nn/im2col_matmul.cc	Replaces local OIHW2OHWI implementation with call to TransposeKernel; adds conv.h include for the function

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[daijh]

Fix the CI error for span.

[guschmue]

[daijh]

Thanks for analyzing the CI failure.

The webgpu_minimal_build_edge_build_x64_RelWithDebInfo environment uses different build options than my local setup.
Added AreSpansEqual to resolve the compiler errors regarding gsl::span comparisons.

[daijh]

React Native CI Pipeline / React Native CI Android (pull_request)

This failure appears to be environment-specific and is unrelated to the changes in this PR.

/usr/local/lib/android/sdk/emulator/qemu/linux-x86_64/qemu-system-x86_64: error while loading shared libraries: libpulse.so.0: cannot open shared object file: No such file or directory