Bug fix for correlated errors and ML
- If a shock variance was set to 0 in calibration, the correlation matrix featured a 0 on the diagonal, leading to wrong estimation results.
- ML and mode-finding did not impose a 0 lower bound for the standard deviations