Add Complete LangGraph + ProllyTree Integration Example with Persistent Memory Workflow

[zhangfengcdt]

Summary

This PR adds a comprehensive example demonstrating how to use ProllyTree as a versioned memory backend
for LangGraph applications, enabling Git-like versioning of AI agent memory with cross-thread
persistence.

Key Features Added

🔄 Real LangGraph Workflow Integration

• StateGraph workflow with proper LLM → Tool → LLM flow
• Conditional edges for intelligent tool execution routing
• Cross-thread persistent memory using scratchpad tools
• Tool nodes for WriteToScratchpad, ReadFromScratchpad, and search functionality

🧠 ProllyTree Memory Backend

• ProllyVersionedMemoryStore implementing LangGraph's BaseStore interface
• Git-like versioning with descriptive commit messages for every memory operation
• Cross-thread data sharing - different threads can read/write shared memory
• Content-addressed storage with automatic deduplication

🤖 Intelligent LLM Integration

• OpenAI GPT-4o-mini integration for natural language understanding
• Automatic fallback to mock LLM when API key unavailable
• Tool binding with proper function calling capabilities
• Error-resistant design that never crashes on missing API keys

Workflow Demonstration

The example demonstrates a realistic research workflow:

1. Thread 1: Writes research data to persistent scratchpad
2. Thread 2: Reads the same data from different thread
3. Thread 1: Continues with additional research updates
4. Thread 3: New user reads latest research data

All memory operations are versioned with Git-like commits showing complete audit trail.