webgpu: merge batchA into M dimension when batchB==1

[xhcao]

When M is small and batchA is large, there are some invalid elements in each tile, merge batchA into M dimesion would reduce the workgroup count.

Description

Motivation and Context

[Copilot]

Pull request overview

This PR updates the WebGPU MatMul implementation to flatten A’s batch dimensions into the effective M dimension when B has no batching (batchB==1), aiming to reduce workgroup overhead for cases with small M and large batchA. It also adds WebGPU-specific regression tests for additional 3D batched MatMul shapes.

Changes:

• WebGPU MatMul: reshape A/B and treat output as {1, batchA*M, N} when batchA != 1 && batchB == 1 (applies to both the generic and Intel subgroup paths).
• Add WebGPU-only MatMul test cases covering 3D inputs with batchA=3, M=2 and N in {3,4}.

Reviewed changes

Copilot reviewed 2 out of 3 changed files in this pull request and generated 2 comments.

File	Description
onnxruntime/test/providers/cpu/math/matmul_test.cc	Adds WebGPU-only test cases for 3D batched MatMul with batchA>1 and M>1.
onnxruntime/core/providers/webgpu/vendor/intel/math/matmul.cc	Extends the Intel subgroup MatMul reshape optimization from M==1 to all batchA!=1 && batchB==1 cases.
onnxruntime/core/providers/webgpu/math/matmul.cc	Extends the generic WebGPU MatMul reshape optimization similarly, flattening batch dims into M when batchB==1.

---

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

[xhcao]

@qjia7 @jchen10 PTAL. Do you think there is necessary to add a threshold as Copilot mentioned? Are there correct issues when changing dispatch geometry here?

[qjia7]

@qjia7 @jchen10 PTAL. Do you think there is necessary to add a threshold as Copilot mentioned? Are there correct issues when changing dispatch geometry here?

No. It should be a general optimization. Just update your comments in code.

[qjia7]

LGTM. Please also update your PR description. Thanks.