softplus support for webgpu

[guschmue]

needed by onnx-community/Falcon-H1-Tiny-90M-Instruct-ONNX

[Copilot]

Pull request overview

Adds ONNX Softplus operator support to the WebGPU Execution Provider, enabling execution of models (e.g., onnx-community/Falcon-H1-Tiny-90M-Instruct-ONNX) that include Softplus nodes on WebGPU.

Changes:

• Register WebGPU Softplus kernels for ONNX opsets 1–21 and opset 22.
• Implement Softplus as a WebGPU unary elementwise shader using a numerically-stable formulation.

Reviewed changes

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

File	Description
onnxruntime/core/providers/webgpu/webgpu_execution_provider.cc	Registers Softplus kernels in the WebGPU kernel registry for opsets 1–21 and 22.
onnxruntime/core/providers/webgpu/math/unary_elementwise_ops.cc	Adds the Softplus WGSL expression and hooks it into the WebGPU unary elementwise kernel macros (versioned + opset 22).

---

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

[xenova]

LGTM ✅

>>> import torch
>>> (torch.log(1 + torch.exp(torch.tensor(20.0, dtype=torch.float32))) - 20.0).item()
0.0
>>> (torch.log(1 + torch.exp(torch.tensor(11.0, dtype=torch.float16))) - 11.0).item()
0.0

in pytorch, error is zero using these approximations in the relevant cases.