log_prob doesn't drop additive constants

Hello,

log_prob states that:

log_prob returns a value (up to an additive constant) the log posterior

and it is my understanding that syntax of the form y ~ dist() drops additive constants. However that doesn't seem to be the case for the examples I have tried

To reproduce:

library(rstan)

writeLines(readLines("propto.stan"))
## ==>
## parameters {
##     real x;
## }
## model {
##     x ~ std_normal();
## }

writeLines(readLines("target.stan"))
## parameters {
##     real x;
## }
## model {
##     target += std_normal_lpdf(x);
## }

propto <- stan("propto.stan", chains = 0)
target <- stan("target.stan", chains = 0)

x <- 1.0

log_prob(propto, x)
## ==>
## [1] -1.418939

log_prob(target, x)
## ==>
## [1] -1.418939

I would expect the propto code to drop the second term in

-0.5 * x^2 - 0.5 * log(2 * pi)
## ==>
## [1] -1.418939

Not sure if this is an rstsan or stan issue. Let me know if there is anything I can do to help - thanks!

rstan: 2.21.2
R: R version 4.0.3 (2020-10-10)
OS: ubuntu 18.04

Start by reproducing the discrepancy with the propto.stan and target.stan examples through RStan's log_prob function. Compare the generated behavior for the sampling statement and the explicit std_normal_lpdf call, then determine whether the fix belongs in RStan or Stan. Done means the documented additive-constant behavior and the reproduced outputs agree.