fix: add adaptive risk agents example with relative-import tests

[pragam-m25]

Thanks for opening a PR! Please click the Preview tab and select a PR template:

• 🐛 Bug fix
• 🛠 Feature/enhancement

Summary

This PR adds a new diagnostic example demonstrating adaptive risk-taking agents.

Each agent:

• chooses between safe and risky actions,
• maintains a short memory of past outcomes,
• adapts its risk preference based on recent experience.

The example is intentionally minimal and uses only core Mesa primitives.

---

Motivation

Many adaptive agent models in Mesa end up concentrating:

• decision-making,
• action execution,
• memory updates,
• and learning logic

inside a single agent.step() method.

This example is designed to make that complexity explicit, rather than hide it, and to serve as a concrete, example-first input for discussions around richer behavioral abstractions in Mesa.

---

Scope

• New example only (no API or behavior changes)
• Uses current Mesa patterns (no deprecated schedulers, no DataCollector)
• Relies on model.random for reproducibility
• Includes minimal smoke tests to ensure the model initializes and steps

---

Testing

• Added smoke tests verifying:
  • model initialization
  • stepping without errors
• Tests intentionally do not assert outcomes or dynamics

---

Notes

This example is exploratory and educational by design. It does not propose new abstractions or patterns, but highlights current modeling constraints through a working example.