allow setting only named parameters

[rmnldwg]

The emcee.EnsembleSampler has the option to provide the parameter proposals as a dictionary with parameter_names as their keys. For sampling, this is useful, as especially the more complex models may have a lot of parameters that we do not always want to set (e.g. noext_contra_IItoIII_spread should be set only via IItoIII_spread most of the times).

For reproducibility purposes, it could be important to allow storing these parameter_names in the model (and its definition). Because unfortunately, the emcee.EnsembleSampler does not store the parameter_names in its backend.

I thought about an additional method, e.g. .set_named_params() that only allows setting parameters that are in a list of parameter_names provided to the constructor of the model. It could still allow setting them via a tuple, which would mean that - as long as the model is reliably reproducible - I can just pass in the samples from a previous run while being sure that the correct value is set at the right place in the model.

This method (and a corresponding .get_named_params()) could be implemented via a simple mixin and it would not introduce any breaking changes, fortunately.