Deal with situations where the Autocorrelation Time is less than the time between thinned samples

[rambaut]

For some well-mixing parameters, the ACT can be less than the time between samples in the log file. This produces wildly inaccurate ESS values which should anyway be bounded at the number of samples in the log file.