Switch to CmdStanR

1. The migration will break how get_mcmc() behaves — currently the function returns a rstan::stanfit object. The rstan methods (summary(), get_stancode(), etc. ) work on this object. I can't find any package's function that converts a CmdStanMCMC object --> rstan::stanfit object and writing this would take a lot of effort. An updated get_mcmc() could return the CmdStanMCMC object, which also has a set of associated methods/functions.

We're leaning toward: Clean break after migration, with error message (get_mcmc() would now return CmdStanMCMC object, error if given a metab_model with rstan::stanfit) and point to the updated docs for how to work with CmdStanMCMC objects.

2. Unlike rstan, CmdStanR doesn't bundle Stan — users need to call cmdstanr::install_cmdstan() separately after installing the R package. At minimum I'd add an error message from metab_bayes() if CmdStan isn't found. The question is whether we should go further, adding a helper function like install_stan() that wraps cmdstanr::install_cmdstan(). This could be called on its own, or conditional in metab_bayes() if CmdStan isn't found, with a message.

We're leaning toward: error message from metab_bayes()

The main entry points are get_mcmc() and metab_bayes(); begin by tracing their current return behavior and CmdStan availability checks. Confirm the clean-break behavior for rstan::stanfit inputs and the missing-CmdStan case, then update the documentation for CmdStanMCMC workflows and verify both error paths.

Written by the indexing model from the issue text.