fix oob read access in GatherBlockQuantized

[guschmue]

GatherBlockQuantized shader could read oob for ill formed indicies.
The correct solution would be to throw and exception but the indicies input sits on gpu and we don't want to take the performance hit by copying it to cpu.
Instead, we only check inside the shader to not read oob.

[Copilot]

Pull request overview

Fixes potential out-of-bounds reads in the WebGPU GatherBlockQuantized shader when indices fall outside the gather-axis range. Since the indices live on the GPU, the check is performed inside the shader (rather than on CPU) to avoid a host copy; invalid indices simply produce zero outputs.

Changes:

• Compute gather_axis_dim once, then bounds-check the (possibly negative-adjusted) index and write 0 + early return on OOB.
• Add a WebGPU regression test exercising both positive (2) and negative (-3) out-of-range indices for int32 and int64 index types on 2-bit/uint8 data.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
onnxruntime/contrib_ops/webgpu/quantization/gather_block_quantized.cc	Adds shader-side bounds check after negative-index normalization; OOB lanes write 0 and return.
onnxruntime/test/contrib_ops/gather_block_quantized_op_test.cc	New WebGpu_InvalidIndices_2Bits_Uint8 test covering invalid indices for both int32 and int64 index tensors.

---

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