-----
* Add local transition class which makes a local KDE fit.
* Fix corner cases of adaptive population size strategy
* Change the default: Do not stop if only a single model is alive.
* Also include population 0, i.e. a sample from the prior, in the websever
visualization
* Minor bug fixes
* Fix inconsistency in ABC options if db_path given as sole string argument
* Add four evaluation parallel samplers
* Dask based implementation
* More communication overhead
* Future executor evaluation parallel sampler
* Very similar to the Dask implementation
* Redis based implementation
* Less communication overhad
* Performs also well for short running simulations
* Multicore evaluation parallel sampler
* In most common cases, where the population size is much bigger
than the number of cores, this sampler is not going to be faster
than the multicore particle parallel sampler.
* However, on machines with lots of cores and moderate sized populations
this sampler might be faster