feat(faz4): Add multi-agent LLM system with orchestrator, planner, ex…

[Kaangml]

…ecutor, evaluator

Multi-agent architecture:

• Orchestrator: Coordinates plan→execute→evaluate loop
• PlannerAgent: DOM-aware multi-step planning with LLM
• ExecutorAgent: Step execution with retry logic
• EvaluatorAgent: Result analysis and re-planning triggers

LLM provider abstraction:

• BaseLLMProvider with Message/LLMResponse dataclasses
• AWS Bedrock provider (langchain-aws)
• Google Gemini provider (langchain-google-genai)
• OpenAI provider (langchain-openai)
• MockLLMProvider for testing without API calls
• Factory for provider creation by role

Browser enhancements:

• DOMAnalyzer for intelligent page structure extraction
• PageStructure model for interactive elements and forms

Configuration:

• .env.example with all provider API key templates
• LLMConfig with load_from_env() support

Testing:

• 68 tests passing, comprehensive coverage for new components
• All tests use MockLLMProvider (no real API calls needed)