feat: flux sampling with optlang solvers (#484) [fix #415]

- port the cobra legacy solver sampling implementation to optlang solvers
- sample now takes a fluxes arg
- validate can validate both sample outputs
- sample and batch now return DataFrames
- updated docs
- Non-homogeneous sampling is now almost as fast as homogeneous sampling
  due to pre-projecting search directions
- ARCHSampler renamed to ACHRSampler as it is known in the literature
- changed internal problem representation
- even less frequent reprojection, method seems to be pretty stable
- sample now also allows setting the thinning parameter