BUG: using the soft_init sampler kwargs with dynesty caused an AttributeError

Closes #690 (closed) I thought that it was not ideal that _log_likelihood_eval_time was only defined "behind the scenes", so I modified the _time_likelihood() function to return the time instead of setting self._log_likelihood_eval_time, so that it can be set explicitly in init

However, looking at the function, it is also clear that the value of the _log_likelihood_eval_time could still be NaN, regardless of the value of soft_init, which is why I also modified the dynesty-specific code to use the default value of n_checkpoint (10) if the eval time is NaN.