Make groom publisher fixture date-relative instead of hard-coded

[qapdex-maker]

The token-failover job of evaluation-workflow-tests went red on
2026-10-02 (run 37046657914) in test_groom_publisher_preserves_resolved_dispatched_rows.

Cause

devops-health-groom.md drops a resolved outbox row once it is older than
14 days, using Date.now():

const resolvedOutboxExpired = prior => {
  ...
  const ageDays = Math.floor(
    (Date.now() - Date.parse(`${date}T00:00:00Z`)) / 86400000
  );
  return ageDays > 14;
};

The fixture in run_groom_publisher_without_rows pinned correlation_date to
the literal 2026-09-16. Once real time passed that date the row was treated
as expired, silently omitted from the rendered dashboard, and the assertion
assertIn("🔄 Dispatched", ...) failed with an empty island table.

This is a time-bomb fixture, not a product bug. It will fail again for anyone
running the suite later than 14 days after the pinned date.

Fix

Default correlation_date to yesterday, so the row is always inside the
retention window, and derive first_seen and the table's First Seen column
from the same value. The expiry test still passes 2000-01-01 explicitly, so
the expiry path stays covered.

Verification

Same worktree, before vs after the patch:

• baseline: FAILED (failures=1, errors=23)
• patched: FAILED (errors=23)

The 23 errors are all FileNotFoundError: 'pwsh' from the local machine having
no PowerShell, unrelated to this change. The single failure this fixes is gone.

[github-actions]

Note

This PR is from a fork and modifies infrastructure files (eng/ or .github/).

Changes to infrastructure typically need to be submitted from a branch in dotnet/skills (not a fork) so that CI workflows run with the correct permissions and secrets.

Please consider recreating this PR from an upstream branch. If you don't have push access to dotnet/skills, ask a maintainer to push your branch for you.

[AbhitejJohn]

Thanks a bunch for catching and fixing this, @qapdex-maker. I preserved your original commit and credit in the trusted follow-up PR #1264. I also added a small follow-up so the status-parser test checks the new date-relative correlation instead of the old fixed value.

(Copilot, commenting on Abhitej's behalf.)