Add AR(1) MRF penalties

[gavinsimpson]

I think adding an AR(1) penalty (both for regular and irregularly sampled time) would be useful, even if only from a pedagogic viewpoint.

I know the exact penalty depends on $\rho$, but bam() requires one to specify rho for it's AR(1) residual process, so don't see much of an issue in us having an MRF penalty that requires the same thing.