Stop the no_std CI job hanging on apt

[richardanaya]

The no_std job on master commit b3c0332 (run 37679549289, job 112991706416) never built or booted QEMU. GitHub cancelled it at the 6-hour limit. The same job passed on PR #17 (67074c3).

Root cause

sudo apt-get update hung. GitHub's Ubuntu image lists mirrors in /etc/apt/apt-mirrors.txt:

1. http://azure.archive.ubuntu.com/ubuntu/ (priority 1)
2. https://archive.ubuntu.com/ubuntu/ (priority 2)

On the cancelled run, Azure ignored InRelease (Ign at 20:07:07Z). Apt then logged Ign for noble-updates package indexes on that same host (20:07:09Z) and printed nothing else until the cancel at 02:07Z. It did not fail over, even though https://archive.ubuntu.com had already returned InRelease in that run. The passing run used Azure too, and Azure answered, so this is the mirror stall, not the harness.

Fix

• Point apt at https://archive.ubuntu.com only, so the broken failover path is not used.
• Set apt acquire timeouts (30s, 3 retries) and kill a still-stuck apt-get after 180s with a message.
• timeout-minutes: 15 on the job.
• Run QEMU under timeout (60s). A guest that never semihost-exits fails immediately instead of holding the runner.

No library changes.

CI

Three separate runs, each with every job green, including no_std build and qemu:

• https://github.com/richardanaya/executor/actions/runs/37718539791 (ff5811c)
• https://github.com/richardanaya/executor/actions/runs/37718830034 (ef300aa)
• https://github.com/richardanaya/executor/actions/runs/37719120991 (8b6ead2)

The first run's apt log fetched only from https://archive.ubuntu.com (mirror list was 46 bytes, not the 144-byte Azure list) and QEMU printed executor nostd tests ok.