Clean up sampling tests (#1486)

* Moved dyntesty mcmc test into pipeline

* Merged two very similar tests

* Remove unused sample result

* Sized down sample size to a fifth (still getting through in 80% of test) and marked as flakey.

* Moved harmonic mean to other evidence calculations

* Added util file for problem definition and constants

* Fixed Small typo and limited the number of chains for pymc

* Removed "test_groundtruth_separated_modes", as it is not really a test as for of an exhibition, which is covered in the jupyter notebook "sampler_study"

* Adapted import in variational to new changes

* Fixed error in multi_startpoint tests due to increased number of chains