Add agent execution tracing and structured logging

[Syedumer-97]

Summary

CortexON currently lacks a unified execution tracing mechanism across agent runs. This makes debugging, observability, and future replay difficult.

This issue proposes adding:

• A run_id per user request
• A step_id per agent/tool execution
• Structured logs correlated via these IDs
• Integration with Pydantic Logfire

Motivation

• Debugging multi-agent workflows
• Observability in production
• Agent behavior introspection
• Foundation for future replay & analytics

Proposed Design

TraceContext

A lightweight context object containing:

• run_id
• agent_name
• step_id
• tool_name (optional)

Logging

• Structured JSON logs
• Every log includes run_id and step_id
• Configurable via ENV

Scope (Initial)

• Backend only
• No persistence (in-memory / logs only)
• No breaking changes

Acceptance Criteria

• Each request generates a run_id
• Each agent action generates a step_id
• Logs are structured and correlated
• Minimal performance impact

Future Work

• Trace persistence
• Frontend visualization
• Execution replay