Allow se() special for gaussian and student families to incorporate sampling error

[nicholasjclark]

{brms} specials are highly useful for many real-world contexts. As a start to incorporating these, the se() might be the most practically useful. {mvgam} currently has no straightforward way to capture this uncertainty (apart from creating pseudo-series that are then forced to share the same process models and observation params, but this won't necessarily let the uncertainty change with time as it should when incorporating known sampling variability)