ci: bound job runtime so wedged runners fail fast

[tobez]

A push run on main recently sat for 15 minutes with all three jobs stuck
(a hosted-runner infrastructure stall — the identical commit passed on
re-run and in its PR run), and without a job timeout such a wedge would
occupy runners for GitHub's default 6 hours before failing.

Normal runs finish in about a minute, so cap the test matrix jobs at
10 minutes and sanitize at 15 (it shows more runner-speed variance
under ASan). A future infra hiccup now self-terminates and reports
failure promptly instead of hanging.

Test plan

• YAML parses cleanly
• The workflow itself runs on this PR — checks going green validates
  the change end-to-end