check_collinearity does not work on a brms model with sampled priors

Hi there,

The check_collinearity function returns a vcov error after passing it a brms model obtained with argument sample_prior = "yes". I checked the help page on check_collinearity, and tested whether vcov works with that model (and it does). Here's a reproducible example:

require(performance)
require(brms)

data("epilepsy")
woprior <- brm(count~Trt*visit + zAge + zBase,
              family = "poisson",
              data = epilepsy,
              seed = 1) 
wiprior <- brm(count~Trt*visit + zAge + zBase,
           family = "poisson",
           sample_prior = "yes",
           data = epilepsy,
           seed = 1)

#check_collinearity throws a vcov error for the second model
check_collinearity(woprior)
check_collinearity(wiprior)

#but vcov works on both models
vcov(woprior)
vcov(wiprior)

The error is:

check_collinearity(wiprior)
Error in stats::vcov(x)[params, params, drop = FALSE] : 
  subscript out of bounds

The only prior being sampled here is the intercept (brms doesn't sample unspecified, improper coefficient priors).

Start by running the reproducible brms example from the issue and compare check_collinearity(woprior) with check_collinearity(wiprior), using vcov output as the reference. Trace the check_collinearity entry point and identify why sampled prior parameters are included in the indexing; done means the sampled-prior model no longer raises a subscript-out-of-bounds error while the non-prior model still works.