Changed signature of reset and sample methods to use a model state dictionary

[luisenp]

Types of changes

• Model.reset() and Model.sample() signature has changed. They no longer receive
  TransitionBatch objects, and they both return a dictionary of strings to tensors
  representing a model state that should be passed to sample() to simulate
  transitions. This dictionary can contain things like previous actions, predicted
  observation, latent states, beliefs, and any other such quantity that the model
  need to maintain to simulate trajectories when using ModelEnv.
• Ensemble class and sub-classes are assumed to operate on 1-D models.

• Docs change / refactoring / dependency upgrade
• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to change)

Motivation and Context / Related issue

This change facilitates some of the work necessary to complete the PlaNet implementation in #86, and should allow more flexibility in the types of models that can be simulated with ModelEnv.

How Has This Been Tested (if it applies)

All unit tests pass and PETS notebook runs correctly.

Checklist

• The documentation is up-to-date with the changes I made.
• I have read the CONTRIBUTING document and completed the CLA (see CONTRIBUTING).
• All tests passed, and additional code has been covered with new tests.