Implements a comprehensive agent memory system for support LLM agents

[zhangfengcdt]

This PR implements a comprehensive agent memory system for support LLM agents with different types of memory persistence backed by prolly tree data structures.

🚀 Key Features

Memory Types Implemented

• Short-Term Memory: Session/conversation history with TTL expiration
• Semantic Memory: Facts, relationships, and structured knowledge
• Episodic Memory: Experience records and interaction history
• Procedural Memory: Rules, procedures, and operational knowledge

Prolly Tree Integration

• Storage Backend: ProllyTree<32, InMemoryNodeStorage<32>> for robust data storage
• Thread Safety: Arc for concurrent async operations
• Advanced Features: Tree statistics, range queries, commit tracking
• Git-like Operations: Branch/checkout/merge simulation for future durability

Also, we converted the financial advisor application to use the Rig framework for LLM interactions while maintaining git-prolly as the storage layer for all objects.

Key Changes

1. Dependencies (Cargo.toml)

• Added rig-core = "0.2.1" dependency for LLM framework integration

2. New Module (src/advisor/rig_agent.rs)

• Created FinancialAnalysisAgent struct that uses Rig's OpenAI client
• Implemented proper Rig agent builder pattern with system prompts
• Added structured request/response types (AnalysisRequest, AnalysisResponse)
• Provides automatic fallback to rule-based analysis when AI fails

3. Refactored FinancialAdvisor (src/advisor/mod.rs)

• Replaced direct HTTP client (reqwest::Client) with FinancialAnalysisAgent
• Updated generate_ai_reasoning_with_debug to generate_rig_analysis_with_debug
• Removed unused imports