Implementation of `matching_variables` and `remaining_columns_to_read` is inefficient and can cause hangs

[sbfnk]

Describe the bug
The draws() function becomes prohibitively slow when used with many variables, e.g. large arrays. This is due to looping over all variables for matching in matching_variables and remaining_columns_to_read from read_cmdstan_csv.

To Reproduce
Example:

library("cmdstanr")

model_code <- '
parameters {
  array[100, 1000] real x;
}

model {
  for (i in 1:100) {
    for (j in 1:1000) {
      x[i, j] ~ std_normal();
    }
  }
}
'

stan_file <- tempfile(fileext = ".stan")
writeLines(model_code, stan_file)

model <- cmdstan_model(stan_file)

sim <- model$sample(chains = 1, iter_sampling = 1, fixed_param = TRUE)

dr <- R.utils::withTimeout(sim$draws(), timeout = 120)

running into a timeout in my instance.

Expected behavior
Ideally: improved run times.

CmdStanR version number
0.5.3