Add function to compute proportion of explained variation

[nicholasjclark]

Using predict() with type = 'terms' on a representative grid of newdata will allow for prediction variance to be decomposed into contributions from unique model terms (i.e. s(cov), s(cov, by = fac)). Can allow for process and observation variance to be calculated separately for deeper insight. Should also give options for a summary() and a plot() function, similar to options for irf() objects