pyGSTi underestimates log-likelihood when data contains 0-counts about pygsti

[enielse]

Cause: This is due to incorrectly dropping logL and chi2 terms corresponding to 0-count (or "zero-frequency") outcomes when these should in fact be nonzero. In the chi2 case, the general term, N(p-f)^2/p, becomes N*p. In the logL within our "poisson picture", the general term N*f*log(p) - N*p becomes -N*p. In both cases these are nonzero even if f==0.

Fix: the hotfix deployed as Release 0.9.7.4 usually fixes this issue with a temporary patch that causes DataSet-generation routines to not omit any probabilities.

[enielse]

Update: a permanent fix that repairs the logL and chi2 functions will be merged into the main branches by pull request #64. We'll plan to close this issue once this appears in a release version of pyGSTi.

[enielse]

As promised, closing this issue now that the version 0.98 release is out, which fixes this issue.