Consider how random effects are fitted

Since there are several ways to include random effects for GAMs, it might be worthwhile thinking about which is best.  A simple random intercept (`s(year, bs = "re")`) is probably what most demographers would do.  A random smoother for each year (`s(log_size_prev, year, bs = "fs")`) is probably the best fit to the data.  Ideas outlined in notes/environmental stochasticity.Rmd