Fix subfit issues (mainly due to the lack of xlev)

[fweber144]

This resolves several bugs related to submodel fits of class subfit, see the message of commits fcf1408 and d08920a. The main problem was that in predict.subfit(), argument xlev of model.matrix() was not used. Unfortunately, deriving these xlevels at fitting time is not really straightforward, which is the reason why this PR adds a bunch of code for deriving these at the beginning of fit_glm_ridge_callback() and search_L1().

All of the issues resolved by commit fcf1408 can be reproduced using the following code:

options(mc.cores = parallel::detectCores(logical = FALSE))
N <- 41L
K <- 5L
K_fac <- 4L
set.seed(457324)
dat <- data.frame(
  y = rnorm(N),
  xcat = gl(n = K, k = floor(N / K), length = N,
            labels = paste0("gr", seq_len(K))),
  xfac = sample(gl(n = K_fac, k = floor(N / K_fac), length = N,
                   labels = paste0("fgr", seq_len(K_fac)))),
  xlog = sample(rep_len(c(TRUE, FALSE), length.out = N))
)
levels(dat$xfac) <- c(levels(dat$xfac),
                      paste0("fgr", (K_fac + 1L):(K_fac + 2L)))
dat$xcat <- as.character(dat$xcat)

library(rstanarm)
rfit <- stan_glm(y ~ xcat + xfac + xlog,
                 data = dat,
                 seed = 1140350788,
                 chains = 1, iter = 500,
                 refresh = 0)

library(projpred)

# debug(fit_glm_ridge_callback)
prj <- project(rfit, solution_terms = c("xcat", "xfac", "xlog"),
               nclusters = 1)

dat_new <- data.frame(
  y = c(-1.8, 0.1),
  xcat = rep(paste0("gr", 2), 2),
  xfac = factor(rep(paste0("fgr", 2), 2),
                levels = c(rev(levels(dat$xfac)), paste0("fgr", K_fac + 3L))),
  xlog = rep(TRUE, 2)
)
# debug(predict.subfit)
prjlp <- proj_linpred(prj, newdata = dat_new)

# debug(search_L1)
cvvs <- cv_varsel(rfit,
                  nclusters = 1,
                  nclusters_pred = 1,
                  seed = 46782345)
cvvs <- cv_varsel(rfit,
                  nclusters = 1,
                  nclusters_pred = 1,
                  refit_prj = FALSE,
                  seed = 46782345)

(All of the calls following a commented debug() call failed before this PR and succeed afterwards.)

For the issue resolved by commit d08920a, see the corresponding commit message for reproducibility instructions.