fix(detector): detect partial self matmul output

[prasannakotyal]

Summary

• add SELF_MATMUL_OUTPUT for custom_kernel = functools.partial(helper) where the helper returns data @ data.t()
• require the partial to wrap exactly one helper with no bound args or kwargs
• keep the helper body compute-free except for the self-matmul return or one local alias

Target

KernelGuard-Red-Submission: 87

Validation

• UV_CACHE_DIR=/tmp/uvcache uv run python -m py_compile kernelguard.py
• functools.partial(_c) sample where _c(data) returns data @ data.t(): classification=hacked, should_filter=true, pattern SELF_MATMUL_OUTPUT
• UV_CACHE_DIR=/tmp/uvcache uv run python ../../kernelguard_bypasses/eval_blue_patch.py kernelguard.py clean fixtures remain should_filter=False
• Input reduction detector passed official eval on PR fix(detector): detect input reduction fake output #253 with TP 20/20, FP 20/20, surgicalness 1.0

[github-actions]

KernelGuard Blue Evaluation

• Evaluation job: 199
• Blue submission: 171
• 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/25383660486

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 This PR is now superseded by the consolidated rule-family implementation in #273, which folds this detector coverage together with the related passing-eval variants.