Improvements for v0.2.0

• Easier access to horizon, batch_size, and feature dims (via properties?)
• move symvar / x0 from getattr to a @property decorated fn
• _pytree_ignore_ might be better for batch_dim/traj_dim than whatever hack struct has?
• better handling of cost_params and step_params? difficult to pass them around
• Instead of using local copy of mpc/sysvars.py we can import opt_tools.casadi_tools.sysvars. The __iter__ should be ported there, goal is to keep this module consistent accross projects

---

• sysvarset
  • re-add reduce methods & tests
  • tree_map_reduce to combine tree_map and replace to cast?
  • pickleable?
  • iterators?
  • to_traj / to_batch?
  • ensure at least 1-d? But also compatible w/ optvarset...
  • [DROP: difficulty implemneting, need unclear] update_defaults? Used in brax_sys, must be a better soln...
• optvarset
  • tests
  • check inheriting from sysvar?
  • check usage if can avoid an object, just functional?
  • Update defaults
  • Initializing with non-defaults should be easier (e.g. skip_sym / x0 for casadi) and unambiguous
• sys
  • rewarite/test w/ new format
  • sys into dclass to separate out Input, State? This might make easier.
• ipopt
• Solvers
  • single shooting
  • multi shooting
  • cem
• [DROP: need unclear, backlog] decorators for batch and traj eval, use in step/output in sys.
• make casting to traj or batch easier... at least check why the tree_map_relaxed is needed, see if can reduce
• Seems better to have explicit creation for batches and trajs... there's been a couple bugs where they have been not as expected, would be better to get an error
• Making derived classes means pickle doesnt work out of the box (dill does though). Would be nice if pickle also worked.
• Check how JAX is doing their types for arrays, somehow the error messages have type float32[25,1]