DSPY Integration

[Avngrstark62]

DSPY Integration

This PR introduces DSPy-based prompt optimization into Bindu, enabling agents to improve their system prompts over time using real user feedback, in a safe and fully reversible way.

Key changes include:

• Persistent storage of agent interactions, feedback, and prompt versions in Postgres.
• Runtime prompt routing with canary-style traffic splitting between active and candidate prompts.
• An offline DSPy training flow (SIMBA optimizer) that proposes new candidate prompts without affecting live traffic.
• A separate canary controller that promotes or rolls back candidate prompts based on observed metrics.
• Strict separation between training, routing, and promotion to ensure safety, measurability, and rollback guarantees.

Overall, this turns Bindu agents into self-improving systems while avoiding online learning, unsafe prompt swaps, or irreversible changes.

---

✅ Checklist clearly

• I've added test cases for this feature
• I've ran the tests and all unit test cases are passing(both the new ones and the already existing ones)
• I've tested everything related to this feature
• I've followed the project's style and linting rules.

---

📌 Related issue clearly

Closes #48