Prediction: introduce lane_aggregating_evaluator that uses symmetric functions to aggregate lane_features and classify what lane the obstacle vehicle will select in the future.

planning: open space add naive linear interpolation inside python wrapper to study infeasible problem issue.

* prepare code for multi-scale camera object detection

* add constexpr for const value

* change yolo detecto config for new multi-scale code

planning: fail pull-over sonner based on path_data, and add a temp stop fence so that adc can be paused at a better position