Add NOPE safety tools (Edge, Ocular, Predicate, Crisis Screen, Evals)

[padolsey]

Adding a few open-source safety tools from NOPE focused on detecting mental-health-crisis and relational-safety signals in human–AI conversations:

• Classification — Edge (fine-tuned Qwen3 risk-type classifier), Ocular (behavioral embedding classifier), Predicate (criterion-conditional classifier on Granite)
• AI for Safety — NOPE Crisis Screen (a Guardrails AI validator; pip install nope-crisis-screen)
• Red Teaming Datasets — NOPE Evals (open blueprints benchmarking how public conversational models behave in crisis/relational contexts)

All are publicly usable today (open weights on HuggingFace, a public container, PyPI, and a public configs repo). Added alphabetically within each section.