[templates] Fix Merge_Release and Translate_Changelog on shell executors

[duckhawk]

Summary

Both templates/Merge_Release.gitlab-ci.yml and templates/Translate_Changelog.gitlab-ci.yml fail on GitLab runners that use the shell executor because the templates assume root/docker semantics:

• apt-get install requires root, which shell executors do not have (Permission denied on /var/lib/apt/lists/lock).
• git config --global writes to the runner user's $HOME (e.g. /opt/gitlab-runner/.gitconfig), which is typically not writable (could not lock config file /opt/gitlab-runner/.gitconfig: Permission denied).

Real log excerpt from a failing merge_and_release job on a shell runner:

$ apt-get update -qq && apt-get install -y -qq git curl jq
E: Could not open lock file /var/lib/apt/lists/lock - open (13: Permission denied)
...
MR merged successfully.
error: could not lock config file /opt/gitlab-runner/.gitconfig: Permission denied
ERROR: Job failed: exit status 1

Note that the image: directive in the templates is silently ignored by shell executors, so the job runs directly on the host as the gitlab-runner user.

Changes

• Replace the unconditional apt-get install with a command -v presence check. Install only when the tool is actually missing and the job runs as root (typical for docker images). On shell executors the required tools (git, curl, jq, python3) are expected to be preinstalled on the host, and the job now fails fast with a clear message instead of a cryptic apt error.
• Drop git config --global. Use repo-local git config inside the clone directory (Merge_Release) or git -C "$CI_PROJECT_DIR" config (Translate_Changelog) so nothing is written to $HOME.

The templates continue to work on both docker and shell executors with no runner-side changes beyond having the standard tooling available on the host.

Backport

Cherry-picked to v12.0 in 1c614a3.

Test plan

• Run merge_and_release job on a shell-executor runner with git, curl, jq preinstalled and confirm it tags and creates a Release without touching $HOME or apt.
• Run translate_and_create_mr job on a shell-executor runner and confirm the translation MR is created.
• Run both jobs on a docker runner (e.g. debian:bookworm-slim, python:3.11-slim) and confirm apt-get install still works for any missing tool.
• Re-run the storage-volume-data-manager MR pipeline that previously failed and confirm the job succeeds.