Update RLM usage to match DSPy 3.1.2 export paths and parameter lists.
The module reference changes from Dspy.Predict.RLM to Dspy.Predict.RLMClass
and the unsupported max_depth parameter is removed from initialization.

History inspection now uses dspy.inspect_history directly instead of
evaluating Python expressions against the LM object. This simplifies
the flagship examples and removes debug output from normal runs.

Developer experience improvements:
- README now documents uv installation as a prerequisite for Python
  package setup.
- Added guidance for Deno installation via asdf with a pinned version
  in .tool-versions for reproducible RLM example runs.
- RLM example prints colored install instructions when Deno is missing
  rather than silently skipping.
- Snakepit configuration is now skipped during mix test to avoid
  starting Python workers unnecessarily.
- Runtime Python selection prefers the Snakepit-managed venv to avoid
  mismatched package installations on clean setups.

Example fixes:
- Multi-hop QA example uses a local capitals map instead of relying on
  the LM to retrieve state capital information.
- GEPA and RLM flagships share a consistent fetch_prompt_history helper
  with proper error tuple returns.

Regenerated DSPy wrappers with SnakeBridge 0.11.0. Updated snakebridge
dependency to 0.11.0.