Improve test coverage for continuous-time hierarchical models + Discretizer

[mattlevine22]

Demonstrating xfailing tests that I would like to work:

Priority level 1:

• seems like Discretizer + Hierarchical is not
  tests/test_plate_vector_initial_mean_continuous_filters.py

Priority level 2:

• make reshaping needs consistent and issue helpful errors. For example, why do we need to broadcast initial_mean across M trajectories, but don't need to do this for the initial_covariance? The reason is a bit subtle.
  'tests/test_desirable_shared_initial_mean_hierarchical.py'
  and
  `tests/test_plate_vector_initial_mean_continuous_drift_diffusion.py'

[DanWaxman]

I agree that almost all of these should work; thanks!