Add complex action dependencies to Sparse likelihoods 111

[ran-wei-verses]

Introduced a new B_factor_control_list variable so that the user can specify the control variables each transition factor depends on using a list similar to the B_factor_list introduced to specify state dependencies.

• Made edits to utils, control, learning, agent
• The shape of B matrices is now defined as: [state_dim] + lagging_state_dims + control_dims. Conditioning on actions can be done easily with B[..., *relevant_action_idx]
• Added checks/assertions in agent.__init__
• Added relevant tests. Fixed a bunch of failed tests due to using != rather than is not in agent.__init__.
• All tests passed except for those in test_mmp.py and test_SPM_validation.py

[dimarkov]

I am closing this as it should have resolved in a different branch.