BUGFIX: fix act-walk caching

@serguei.ossokine pointed out that the act-walk is not working with parallelization, the reason for the issue is that the object gets pickled at each call to the sampler and so instance-level attributes aren't being preserved properly.

To fix this, I made the _cache a class-level attribute. I also made sure that the cache is wiped at the beginning of sampling in case people run multiple analyses with this method using a single script/session in a jupyter notebook.

I removed the use of premade instances of the walking methods as it wasn't achieving anything.

EDIT: there was an equivalent problem with the rwalk method that had a similar solution