fix(spurctld): reject zero partition default time

[jamesETsmith]

Motivation

Prevent jobs submitted without an explicit time limit from receiving a zero-length limit when their partition configures DefaultTime=0 . This matches Slurm behavior by rejecting the submission instead of accepting a job that may be terminated immediately.

Fixes #759.

Technical Details

• Changed partition time-limit defaulting to return a submission error when any selected partition explicitly defines DefaultTime=0 .
• Preserved existing behavior for positive defaults, unset defaults, cluster fallbacks, multi-partition requests, and jobs with an explicit time limit.
• Added regression coverage for both the defaulting helper and the complete job submission path.

Test Plan

• Run focused unit tests for partition time-limit defaulting.
• Run the submission-path regression test for DefaultTime=0 .
• Verify Rust formatting.

Test Result

• cargo test -p spurctld apply_default_time_limit : 11 passed.
• cargo test -p spurctld submit_rejects_zero_partition_default_time_when_time_is_unset : 1 passed.
• cargo fmt --all -- --check : passed.

Submission Checklist

[X] Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.
md#pull-requests.

[yansun1996]

Nice, focused fix — this correctly matches Slurm's _valid_job_part semantics (reject on any requested partition having an explicit DefaultTime=0, distinct from the unset/None case and the cluster-fallback-0 sentinel). Traced the call site and confirmed the new fallible path runs before job-ID allocation and the WAL/Raft append, so a rejection here is cheap and side-effect-free. Both new tests exercise the real production entry points and would fail against the pre-fix code.

Two things to handle before merge:

1. This PR currently has a merge conflict with main — could you rebase and resolve it?
2. While you're in there, would it be worth adding a test case where a job requests multiple partitions (e.g. "gpu,cpu") and only one has DefaultTime=0? The current tests all use a single matched partition, so the "reject if any requested partition is zero" semantics (which is the Slurm-faithful behavior here) isn't locked in against a future refactor of the .find() call.

[yansun1996]

@jamesETsmith please resolve the conflicts and amend the above suggested test coverage enhancement.

[jamesETsmith]

@jamesETsmith please resolve the conflicts and amend the above suggested test coverage enhancement.

Thanks for the review, I'll push the changes shortly

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #764      +/-   ##
==========================================
+ Coverage   79.45%   79.48%   +0.03%     
==========================================
  Files         181      181              
  Lines       84439    84504      +65     
==========================================
+ Hits        67088    67162      +74     
+ Misses      17351    17342       -9     

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.