Expand observability events to capture LLM input/output per turn

[astrobot-houston]

Originally proposed by @lforst in #173. Their description is reproduced verbatim below.

All community-submitted pull requests are automatically converted to issues (bugs) & discussions (feature requests, enhancements) where they can be triaged and prioritized. Once prioritized, a PR implementation is created automatically.

Adds an turn_start observability event emitting provider, model, inputs (system prompt, messages, tools) and reasoning level. Additionally expands the turn event to include used provider, LLM output

Having such an event is useful for AI observability, helping understand exactly what the LLM is thinking between turns.

---

Implementation wise, the typings for toTurnMessage and toTurnContent which are new mapping functions, turning Pi output into Flue event data, are a bit ugly.

The other thing I am not 100% about is the new private activeTurnId field on Session. It relies heavily on the fact that there con only be one active turn for a given session. I don't know if that might become a problem in the future.

---

Original implementation from #173 by @lforst