Evaluate timeout records

[katstankiewicz]

• Set a max number of attempts where a conversation reaches conversation timeout
• Set a default of 1 because this is a model error and we want model stats based on this run, ie task_completion, conciseness etc.
• Sets a hard cap for conversation length for models to adhere to
• User behaviour and speech fidelity must still be 1.0 to get evaluated
  • If max_timeout_attempts is > 1, keep a cache of any previous runs that had timeout reached and user_behaviour == 1.0 so we can evaluate that run instead of trying again after meeting max_timeout_attempts
    • eg. max_timeout_attempts == 2, first run timesout and User behaviour == 1.0 but second run also timesout and user behaviour is 0. In this case don't run again, take the first run
• Add conversation_finished_on_time as a diagnostic metric to see how many finished on time
• Save timeout_accepted in result.json to show when a conversation timedout but we evaluate it
• Update user behaviour fidelity metric to handle four conversation end modes: error, timeout, inactivity timeout and user ended call
• Update default conversation timeout to 10 minutes (600s)