Lua unit jobs fail when the www.lua.org download takes over 30 s #92

Summary

The ci unit jobs for Lua 5.1 to 5.4 download the Lua source tarball from www.lua.org on every run, in one attempt with a 30-second timeout and no cache. When the site is slow the job fails in setup, before any test runs. libtmux-lua#89 failed this way twice in a row on 2026-11-05: Unit Lua 5.2 on the first attempt and Unit Lua 5.4 on the rerun.

Reproduction

Steps as reported; not re-run (it depends on www.lua.org answering slowly):

1. Any ci run whose download from www.lua.org takes longer than 30 seconds, for example run 42005144606, Unit Lua 5.4.

Expected

A slow or failed download is retried, or served from a cache, before the job fails.

Actual

  File "/home/runner/work/libtmux-lua/libtmux-lua/scripts/bootstrap_runtimes.py", line 67, in main
    with urllib.request.urlopen(f"https://www.lua.org/ftp/{archive.name}", timeout=30) as response:
TimeoutError: timed out
urllib.error.URLError: <urlopen error timed out>
##[error]Process completed with exit code 1.

Evidence

scripts/bootstrap_runtimes.py fetches the tarball once when .cache/toolchains/lua-<version> holds no build:

        if not archive.exists():
            with urllib.request.urlopen(f"https://www.lua.org/ftp/{archive.name}", timeout=30) as response:
                archive.write_bytes(response.read())

.github/workflows/ci.yml runs it in every unit job and restores no .cache/toolchains from a cache, so every job starts with no build and downloads again.