Figure Out How $t_0$ Should Be Handled

[DanWaxman]

Right now, we assume an observation at t_0 = obs_times[0]. This is potentially unintuitive, especially for continuous-time models. We should probably instead

1. Infer t_0 = obs_times[0] if and only if it is not specified
2. Properly compute the transition $p(x(t_1)) = \int p(x(t_1) | x(t_0)) p(x(t_0)) , dt$ if $t_0$ is specified and differs from t_1 := obs_times[0].
   a. This is a bit different for continuous and discrete time, but should be done for both (and for discrete time, will depend on how Figure Out How Times Are Handled in Discrete Time #206) is handled.

[mattlevine22]

We want this to work across hierarchies/trajectories