Add weights for model fitting

[samuel-watson]

This is essentially the same as adding the heteroscedastic errors (issue #2 ) by adding the weights to matrix W.