stan_lm silently ignores autoscale argument if prior_intercept is normal

Summary:

stan_lm() (really stan_biglm.fit and everything it's used for) seems to silently ignore the autoscale argument to normal() when used for prior_intercept. @bgoodri Is ignoring autoscale a bug or intended? I guess regardless of whether it's a bug or not, I don't think it should be silent.

Reproducible Steps:

# fit same model with same seed, only difference is autoscale argument
fit <- stan_lm(mpg ~ wt + qsec + am, data = mtcars, 
               prior = R2(0.75), prior_intercept = normal(0, 10, autoscale=TRUE),
               chains = 1, iter = 500, seed = 12345)
fit2 <- update(fit1, prior_intercept = normal(0, 10, autoscale=FALSE))

# draws are identical
identical(as.matrix(fit1), as.matrix(fit2)) 
[1] TRUE

RStanARM Version:

2.18.9

R Version:

3.5.2

Operating System:

OS X 10.12.6

Start with the stan_lm() and stan_biglm.fit paths that process prior_intercept and the normal() autoscale argument. Reproduce the two fits from the issue, then trace why their draws are identical and determine whether the intended fix is to apply autoscale or report that it is ignored; add a regression test for the chosen behavior.