[BUG] Dataset downloader reports success and replaces data after a failed download #2276

Describe the bug

datasets/get_test_data.sh can return exit status 0 after a download fails, while removing and recreating an existing destination directory. A developer can therefore lose the prior dataset and receive a success status even though the replacement was never downloaded or extracted.

Steps to reproduce

From a source checkout, this uses a temporary directory and a fake wget; it does not touch an existing dataset or make a network request:

task_dir=$(mktemp -d)
mkdir -p "$task_dir/bin" "$task_dir/work/cvrp"
cp datasets/get_test_data.sh "$task_dir/work/"
printf 'existing data\n' > "$task_dir/work/cvrp/sentinel"
printf '#!/bin/sh\nexit 8\n' > "$task_dir/bin/wget"
chmod +x "$task_dir/bin/wget"
(
  cd "$task_dir/work"
  env -u CUOPT_S3_URI -u CUOPT_AWS_ACCESS_KEY_ID -u CUOPT_AWS_SECRET_ACCESS_KEY \
    PATH="$task_dir/bin:$PATH" ./get_test_data.sh --cvrp
  printf 'exit status: %s\n' "$?"
  if test -f cvrp/sentinel; then echo 'original preserved'; else echo 'original removed'; fi
)

Observed: the script reports a failed download and missing archive, prints exit status: 0, and prints original removed.

Expected behavior

Report a nonzero status when a selected dataset cannot be downloaded or extracted, and preserve an existing destination unless a complete replacement is ready. The same distinction should hold for any selected dataset, not just CVRP.

Environment and scope

Reproduced on Linux against main at 52d14b4c97de5591a8054ef1971a92e10617aed7. The fake-download case exercises the HTTP fallback in a disposable directory. An isolated ordinary run also observed an HTTP 404 followed by status 0; the issue is the script's failure handling rather than the availability of any particular external URL.