ci: build-native-matrix build job has no timeout-minutes — an apt-get stall burned 6h of runner time

[carlos-alm]

What happened

On PR #2586, build / Build x86_64-unknown-linux-musl (job log) ran for 6h 0m 51s before GitHub killed it with The operation was canceled. It never got past the first line of the "Install cross-compilation tools (x86_64-musl)" step:

06:54:11Z  Run sudo apt-get update
06:54:42Z  Ign:2 http://azure.archive.ubuntu.com/ubuntu noble InRelease
06:54:49Z  Get:5 https://archive.ubuntu.com/ubuntu noble-security InRelease [126 kB]
12:54:47Z  ##[error]The operation was canceled.

apt-get update hung against the Azure Ubuntu mirror (note the Ign: retries against azure.archive.ubuntu.com) and never returned. The job sat idle for six hours, failed the PR, and consumed a runner slot the whole time.

Root cause

The build job in .github/workflows/build-native-matrix.yml declares no job-level timeout-minutes, so it inherits GitHub's 360-minute default. Individual steps are mostly unbounded too — only "Install napi-rs CLI" has timeout-minutes: 5. This is inconsistent with ci.yml, where nearly every step carries an explicit 5–30 minute bound.

Because build-native-matrix.yml is a reusable workflow shared by build-native.yml (PR validation) and publish.yml (release builds), the same unbounded stall can also hang a release.

Suggested fix

• Add a job-level timeout-minutes to the build job sized to the real build time (the passing targets in that same run took 1m44s–7m10s, so ~30 min is generous headroom).
• Bound the apt step specifically, and consider making the mirror stall self-healing rather than fatal — e.g. timeout-minutes on the step plus a retry, in the spirit of the existing 3-attempt npm install retry in ci.yml.

Impact

Infrastructure only — no product bug. It converts a transient mirror outage into a 6-hour red PR instead of a fast, retryable failure.