[Test] Fail fast in smoke wait-loops on terminal job failure

[kevinmingtarja]

Summary

The job wait helpers (get_cmd_wait_until_*_job_status_contains_*) poll sky (jobs) queue until the target status appears or the timeout elapses. If a job reaches a terminal failure status (FAILED, FAILED_SETUP, FAILED_PRECHECKS, FAILED_NO_RESOURCE, FAILED_CONTROLLER, CANCELLED) it can never reach a target like SUCCEEDED, yet the loop keeps polling for the full timeout (often 10 min), wasting minutes per failing test.

Add an early exit: when the matched job is in a terminal failure status that is not one of the statuses being waited for, print the status + queue and exit non-zero immediately. Tests that legitimately wait for a failure status are unaffected — the target status is excluded from the fail-fast set.

Test plan

• Generated the wait command for a job waiting on SUCCEEDED and ran it against captured sky jobs queue output:
  • Job in a terminal failure status → exits non-zero immediately instead of polling until timeout.
  • Job in SUCCEEDED → reports target reached, exits 0.
  • Job still running → keeps polling (unchanged).
• Verified a helper waiting for a failure status (e.g. FAILED) excludes that status from the fail-fast set, so it still waits as before.
• All four get_cmd_wait_until_* variants format without error; yapf/isort clean.

---

Part of a 3-PR series cleaning up the smoke-test failure path:

• [Test] Match job by name in any column in smoke wait-loops #9999 — match job by name in any column
• [Test] Fail fast in smoke wait-loops on terminal job failure #9996 — fail fast on terminal job failure (this PR)
• [Test] Bound failed-job log fetch on failure teardown #10000 — bound failed-job log fetch on teardown

[gemini-code-assist]

Warning

Gemini encountered an error creating the review. You can try again by commenting /gemini review.