Reasoning Trace Logger

Implement a reasoning trace logger for debugging LLM chain decisions.

## Chain Components
{{chain_components}}

## Debugging Requirements
{{debugging_requirements}}

## Storage Backend
{{storage_backend}}

Build the logger:

```python
class ReasoningTraceLogger:
    def start_trace(self, request_id: str, input_data: dict) -> TraceContext:
        """Initialize new trace"""
        pass
    
    def log_step(self, ctx: TraceContext, step_name: str, input: Any, output: Any, metadata: dict) -> None:
        """
        Log:
        - Step inputs/outputs
        - Latency
        - Token usage
        - Model used
        - Intermediate reasoning
        """
        pass
    
    def log_decision(self, ctx: TraceContext, decision: str, alternatives: List[str], rationale: str) -> None:
        """Log decision points"""
        pass
    
    def export_trace(self, request_id: str, format: str) -> str:
        """Export for analysis"""
        pass
```

Include:
- Structured logging format
- Trace visualization
- Search and filtering
- PII redaction