BUG: fix dynesty crash when autocorrelation length estimate and/or nact is very long

Closes #668 (closed) In the dynesty sampler loop, return the current live points when the number of "found" points is smaller than the estimated autocorrelation length (times the thinning factor). This prevents a divisionbyzero error since in that case, thinning is not applicable anymore.

Would it be worth adding a logger message that recommends lowering the "nact" value?