Detect rank-deficient matrix in adaptive covariance MCMC

[ben18785]

When models are close to unidentified, in adaptive covariance MCMC, the covariance matrix can become rank-deficient. If that occurs, the linear algebra used to form it may fail and halt sampling. We should instead include a try/catch condition to these routines whereby if it fails we stop adapting and use the previous covariance matrix from then on.