Fast calculation for large SN likelihoods

[joezuntz]

Rick Kessler suggested a speed trick that's useful for many-sample SN likelihoods - calculating a diagonal covariance likelihood first and only if it is below than a certain threshold using the full dense matrix.