CSTR 1.5 punchlist

[scotstan]

Using the pythonsim branch for now.

• [x] Single conda environment.yml file on python 3.10.1 for all
• [x] Add new CSTR simulator running in Python (to parallel the Simulink version)
• [x] Linear and Non-linear MPC controllers for baseline on Python sim
• [x] Spec for common results.csv file to compare
• [ ] Strategy to train four distinct brains to support: modular-, modular-brain-ml, modular-brain-mpc, monolithic-brain
• [ ] Choose graphics path: GIF from do_mpc vs animation using matplotlib
• [ ] Decouple run-time graphics from results generation (for batch scripts)
• [ ] Batch drive the entire benchmark suite
• [ ] Test against CSTR 1.0 suite
• [ ] Parameterize --noise=0.01 or use ENV across the samples

[scotstan]

Current results. Linear MPC fails all cases at 10% error (expected). Can handle 1% noise.