The computations of rules for ContinuousTransition node can be improved

[albertpod]

Following the discussions during PR, @bartvanerp has identified a few places where the computations around the ContinuousTransition node can be improved. In particular

1. If the transformation of RV a is linear, then the computation of Jacobians can be done only once.
2. There are multiple places where trace operation within ContinuousTransition can be optimized, e.g., in the computation of the average energy.