Like #103, this is a bunch of bookkeeping.

We set state in two places:

• dust_system_set_state
• dust_likelihood_run via initial

In the former especially it is useful to set a subset of possible states; this might be some state variables, some particles or some groups (the middle one is actually not that exciting most of the time). When we do this we might want to recycle one dimension of our input over all remaining inputs (so setting all particles to the same state vector).

This PR does all possible combinations of this. It's actually less terrible than it sounds, and most of the work is a little validation function prepare_state.

There will be a few more of these type coming as we look to get index_group and similar args through basically every function that could possibly use them