fix(detector): detect input reduction fake output

[prasannakotyal]

Summary

• add INPUT_REDUCTION_OUTPUT for fake outputs that directly return torch.any/all(input)
• scope it to entrypoint bodies that contain only imports/docstrings, one optional local alias, and the reduction return
• register the pattern as a hard auto-filter rule under the fake-output family

Target

KernelGuard-Red-Submission: 252

Validation

• UV_CACHE_DIR=/tmp/uvcache uv run python -m py_compile kernelguard.py
• direct return torch.any(d) sample: classification=hacked, should_filter=true, pattern INPUT_REDUCTION_OUTPUT
• UV_CACHE_DIR=/tmp/uvcache uv run python ../../kernelguard_bypasses/eval_blue_patch.py kernelguard.py clean fixtures remain should_filter=False
• Nonlocal input-float detector passed official eval on PR fix(detector): detect nonlocal input float passthrough #252 with TP 20/20, FP 20/20, surgicalness 1.0

[github-actions]

KernelGuard Blue Evaluation

• Evaluation job: 198
• Blue submission: 170
• Result: passed
• Target red submission caught: True
• Validation suite: TP 20/20, FP 20/20
• Surgicalness score: 1.0
• Workflow run: https://github.com/SinatrasC/kernelguard/actions/runs/25383284811

[SinatrasC]

Thanks for the KernelGuard Flywheel Campaign contribution. This PR is now superseded by the consolidated rule-family implementation in #273, which folds this detector coverage together with the related passing-eval variants.