Use MPLE predictor matrix to detect extreme statistics

[krivit]

For any dyad-independent statistic, one can obtain its maximum and minimum values under the sample space from the MPLE predictor matrix (provided it's a dyad census) by taking the weighted sum of the positive change statistics and the sum of the negative change statistics.

This can be used to find "dropped" terms even when the InitErgmTerm* function doesn't provide its maxima and minima.