[Fix] Fix reduction fp16 exceed

[hanrui1sensetime]

Thanks for your contribution and we appreciate it a lot. The following instructions would make your pull request more healthy and more easily receiving feedbacks. If you do not understand some items, don't worry, just make the pull request and seek help from maintainers.

Motivation

RTMPose FP16 model will be nan on Android platform using ncnn. So force set ncnn Reduction op for FP32.

Modification

Add 31=15 to Reduction layer params automatically when executing onnx2ncnn.

BC-breaking (Optional)

None.

Use cases (Optional)

None.

Checklist

1. Pre-commit or other linting tools are used to fix the potential lint issues.
2. The modification is covered by complete unit tests. If not, please add more unit tests to ensure the correctness.
3. If the modification has a dependency on downstream projects of a newer version, this PR should be tested with all supported versions of downstream projects.
4. The documentation has been modified accordingly, like docstring or example tutorials.

[grimoire]

LGTM

[nihui]

这个31参数feature是怎么被发现的

[lvhan028]

这个31参数feature是怎么被发现的

正如 @hanrui1sensetime 在motivation中写的那样：
RTMPose FP16 model will be nan on Android platform using ncnn

FYI, RTMPose 模型还没有正式release