feat(chaos): add ChaosMonkey fixture and pipeline resilience tests

[purvask2006-collab]

🔗 Related Issue

Closes #<! -- 144 -->

📝 Summary of Changes

Implements chaos engineering tests to verify that ExecraPipeline continues operating correctly when individual subsystems fail randomly.

Adds a ChaosMonkey fixture that injects three independent fault types:

Subsystem	Fault	Probability
LLM client	TimeoutError on complete()	20%
Redis	None cache miss on get()	50%
Perception queue	Silent frame drop on put()	30%

The pipeline runs for 60 seconds under all three faults simultaneously. On every fault it falls back gracefully — delivering degraded guidance (confidence < 0.5) rather than crashing. All faults are captured in a structured error_log.

Also adds docs/resilience.md documenting each failure mode, recovery strategy, and how to extend the suite.

🔍 Type of Change

• 🧪 Test addition or improvement
• 📖 Documentation update

🧪 How Was This Tested?

pytest tests/chaos/test_pipeline_resilience.py -v --timeout=120

15 tests covering:

• 60s chaos run — no unhandled exceptions
• Guidance delivered on ≥ 75% of ticks under combined chaos
• Every injected TimeoutError logged with kind == "llm_timeout"
• Statistical rate checks within ±15% of configured probabilities
• Fallback guidance carries confidence < 0.5
• Combined/cascading fault scenarios
• ChaosMonkey counter accuracy

Environment: Windows 11 · Python 3.14

✅ Pre-Submission Checklist

• Branch is up to date with upstream/main
• Docstrings and type hints on all new public functions
• Documentation updated (docs/resilience.md)
• PR title follows Conventional Commits format
• No .env files, secrets, or model weights committed
• PR scope matches linked issue only

💬 Notes for Reviewer

ChaosMonkey uses random.Random(seed=2026) — fault sequence is fully reproducible across CI runs.

ExecraPipeline is a thin orchestration wrapper — zero changes to existing production code.

Requires pip install pytest-timeout for the 120s CI timeout guard.