Support compression in read_cmdstan_csv

Cmdstan output files can get pretty large. data.table::fread used in read_cmdstan_csv and read_csv_metadata supports gzip and bzip2 compressed files if the R.utils package is installed.

On the cmdstanr side this would require at least modifying the suffix assertions (assert_file_exists) in both functions. And, switching out grep in favor of zgrep to handle the commented headers with the caveat (at least on linux) that the argument order be switched so that --color=never comes before the regex pattern.

So in read_cmdstan_csv:

fread_cmd <- paste0("zgrep --color=never -v '^#' '", path.expand(output_file), "'")

read_metadata_csv:

fread_cmd <- paste0("zgrep --color=never '^[#a-zA-Z]' '", path.expand(csv_file), "'")

and similarly for the WSL specific code.

Start at the read_cmdstan_csv and read_csv_metadata entry points, including their WSL-specific code, and inspect the existing assert_file_exists and grep command construction. Support gzip and bzip2 paths while preserving commented-header handling, then verify both functions work across the described command variants.