[tflite-micro] Add signal kernel fuzz targets for OOB/div-by-zero vulnerabilities

[foodlook]

Summary

Add OSS-Fuzz project configuration and 6 fuzz targets for tflite-micro signal processing kernels.

These targets exercise the pure signal/src/ functions (OverlapAdd, RFFT, Window, Energy, SpectralSubtraction, PCAN) with attacker-controlled size/index parameters that are not validated by the kernel Prepare functions. All 6 targets produce ASan heap-buffer-overflow crashes within seconds.

Motivation

Reported as Google Issue Tracker #523091874. The reviewer requested OSS-Fuzz reproduction format before reward eligibility.

Files

• project.yaml — project registration (C++, ASan+UBSan, libFuzzer)
• Dockerfile — build env (tflite-micro clone + third-party deps)
• build.sh — builds 6 fuzz targets linking real signal/src/*.cc
• 6 signal_*_fuzzer.cc — one per vulnerable kernel

Testing

All targets locally verified with silkeh/clang:latest Docker + ASan + libFuzzer. Crash reproducers are 4-6 bytes each.

[github-actions]

foodlook is integrating a new project:
- Main repo: https://github.com/tensorflow/tflite-micro
- Criticality score: 0.48710