[feat] Improve confidence estimators

• Clean/Refact confidence estimators
• Add failure detector fusioning confidence estimators, 3 errors are tracked :
  • Missing degree of liberty case : check position covariance + motion corruption OR missing edges
  • Divergence, big failure case : check repeated motion corruption or repeated lack of matches
  • Bad local minimum : check if there is an overlap gap and a motion corruption
• In wrappings :
  • If a failure is detected, a recovery mode is enabled with parameters tuned for big motion and map are fixed.
  • Then, the overlap and position covariance are checked to decide if the pose has been retrieved
    • If the pose has been retrieved, the recovery mode is turned off and the old parameters are resetted
• Some commits refer to minor fixes/refact

Here is a demo video of LV working.