consider removing the option to split dates in bayes

because not splitting is faster, simpler, and the only way to go for hierarchical models.

``` r
> sp <- function(split_dates) { replace(
+ specs(mm_name('bayes', err_proc_iid=FALSE),
+ n_cores=3, n_chains=3, burnin_steps=300, saved_steps=200, verbose=FALSE),
+ 'split_dates', split_dates
+ ) }
> dat <- data_metab('10', res='30')
> nosplit <- metab(sp(split_dates=FALSE), dat)
> split <- metab(sp(split_dates=TRUE), dat)
> get_fitting_time(split)
user system elapsed
7.41 3.85 43.85
> get_fitting_time(nosplit)
user system elapsed
1.37 0.42 7.11
```