Context Window Manager

Implement a smart context window manager for LLM applications.

## Model Context Limit
{{context_limit}} tokens

## Content Types
{{content_types}}

## Priority Rules
{{priority_rules}}

```python
class ContextWindowManager:
    """
    Implement:
    - Token counting per content type
    - Priority-based content selection
    - Compression strategies
    - Overflow handling
    - History management
    """
    
    def build_context(
        self,
        system_prompt: str,
        retrieved_docs: List[Document],
        conversation_history: List[Message],
        user_query: str
    ) -> str:
        # Fit everything within context limit
        pass
```

Include:
- Sliding window for history
- Document summarization triggers
- Token budget allocation
- Metrics for context utilization