Issue #, if available:

Description of changes:

• When can_refit_full tag was added, I forgot to add the tag to linear models. This meant we were skipping refit and using the first fold model, which gives slightly worse results that doing a proper refit.
• This PR enables LinearModels to do the refit.
• Because we don't yet use LinearModels in default portfolios, this doesn't impact the normal user experience.

Example on AdultIncome dataset:

Mainline:

                      model  score_test  score_val eval_metric
0          LinearModel_FULL   -0.352494  -0.326698    log_loss
1               LinearModel   -0.352494  -0.326698    log_loss

This PR:

                      model  score_test  score_val eval_metric
0          LinearModel_FULL   -0.351792        NaN    log_loss
1               LinearModel   -0.352380  -0.326852    log_loss

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.