ci: fail the rpm build on an unhandled matrix arch

[prql-bot]

The Set variables step in build-rpm-package logs ::error::Unknown architecture for an unrecognized matrix.target but doesn't exit, so the step still succeeds. ::error:: only emits an annotation — it doesn't fail a step. The job then continues with package_arch never written to $GITHUB_ENV, and jiro4989/build-rpm-action receives arch: "", so a release could ship an RPM built under whatever arch the action defaults to rather than failing loudly.

The sibling build-deb-package job directly above has the identical if/elif/else shape and does exit 1 in the else branch (release.yaml#L163-L171) — this restores that symmetry.

Not reachable today: the rpm matrix carries only x86_64-unknown-linux-musl, since the aarch64 entry is commented out pending jiro4989/build-rpm-action#6. The branch matters when that entry comes back or a new target is added — the point of the guard is to catch a target that no branch handles.

No regression test: this is a one-line guard in a workflow YAML on a branch that requires an unhandled matrix entry to reach, and the repo has no harness for executing release-workflow shell steps. Failure would be observed as the job going red instead of silently mislabeling the package.

Found by the nightly code-quality survey while reviewing .github/workflows/release.yaml.

[prql-bot]

CI note, so this isn't misread as a regression from the change: nightly / nightly-release / build-python-wheels (ubuntu-24.04, prqlc-python, x86_64) failed before running any of this repo's code. The runner couldn't fetch the action tarball:

##[error]An action could not be found at the URI 'https://codeload.github.com/PyO3/maturin-action/tar.gz/e83996d129638aa358a18fbd1dfb82f0b0fb5d3b'
##[error]Failed to download archive '...' after 1 attempts.

That's the Download action repository 'PyO3/maturin-action@v1' step of the runner's action-setup phase — it aborts before ./.github/actions/build-python executes, so nothing in the checkout is involved. The sibling legs using the same action on the same commit (windows-latest, x86_64 and ubuntu-24.04, source) both passed, and the diff here is a single exit 1 inside build-rpm-package, a job the python matrix doesn't depend on. githubstatus.com reported no open incident at the time, so this reads as an isolated codeload fetch flake rather than a broader outage.

nightly is in check-ok-to-merge's needs list, so that required check will go red until the leg is rerun. I couldn't rerun it from this session — gh run rerun --failed refuses while the parent run is still in progress ("This workflow is already running"), and the remaining nightly jobs (cargo-bench, time-compilation (false), the multi-arch builds) run well past this session's polling budget. A gh run rerun 31777923638 --failed once that run settles should clear it.

Everything else that reached terminal on this commit passed. Unverified at the time this was written, all still in progress: cargo-bench, time-compilation (false), several nightly-release build legs, check-links-markdown, measure-code-cov, build-prqlc-c (macos-15), and the check-ok-to-merge omnibus (required) which hadn't registered yet.