@cderv, thank you for your detailed explanation. Now, I understand that it is not trivial (possible) to have nice syntax highlight.
On the other hand, still I consider that the eval=FALSE would make sense for the Stan chunks.
Rationale: the Stan code chunk is translated to a nicely formatted text (without syntax highlight), but the Stan is compiled again and again when knitting, which can be time consuming. For this reason I fitted in advance my Stan models separately and I was planning to re-read them in the Rmarkdown file. I would like to have some nice output when showing/including the Stan code (not cat). The other reason, that the RStudio Desktop is quite frequently crashing (bomb icon) when running Stan code from R.