Support sending traces to MLFlow

### Description

A lot of the agentic libraries (most built in Python, e.g. langchain or dspy) support pushing the whole session contents to MLFlow, which allows for a nicer debugging experience when trying to understand the agent thinking process, the tools it called, etc.

Consider adding something like this to `fantasy`.

The issue names no files, tests, or entry points. Start by locating how fantasy represents agent sessions and traces, then determine the integration boundary; done should mean the complete session can be sent to MLFlow for debugging, with coverage for the supported behavior.

Written by the indexing model from the issue text.