Use angle intervals, not time intervals for better inference accuracy.

[lionfish0]

Currently the training set in the rejection sampling consists of observations with the recorded time intervals: it would do better if it had the reported angle intervals.