Implement a new model validation method: The predictive distance metric

[chenyangkang]

Try to use a predictive distance metric calculated from banding & tracking data as model selection criteria, likely a better alternative to man-made objective function.

Expected changes in BirdFlowR:

1. Overall distance metric calculation functions (parallel to file interval_log_likelihood.R).
2. Distance metric by time elapse. Distance_metric ~ time_elapse regression, regression intercept as a metric? The time_elapse should also be stratified by season.