parameter-estimation

See #413

Suppose one chain is stuck on one mode; another on another mode. If those two chains may sample independently from each mode, the ESSs will be high when, really, they should be near zero since the samples don't represent anything like independently from the overall distribution. This is why multichain ESS makes sense and we should implement it. I feel like this will give a much better picture of t

After a short chat at COMBINE conference, I think it would be very helpful to underline (either in the docs or somewhere else) that pyPESTO doesn't have to use AMICI, but is basically agnostic about the objective function structure.
One could add an example, which is a bit more complex than the Rosenbrock function. That might help users, who want to use pyPESTO e.g. with another simulator.
What

CLI already lints, but programmatic interface doesn't lint by default. Model linting should definitely be performed when import fails to avoid having to reimplement validation checks in amici and provide helpful error messages.

https://jvi.asm.org/content/84/13/6687.long

SBML: https://www.ebi.ac.uk/biomodels/BIOMD0000000546

Estimate of complexity: low