optimize: stop paginating listWorkflowRuns once 24h cutoff is reached

[Copilot]

The daily AIC guardrail was fetching up to 10 pages of workflow runs even when earlier pages already contained runs past the 24h window. Since listWorkflowRuns returns runs newest-first, encountering a stale run means all remaining runs on that page and every subsequent page are also outside the window.

Changes

• Early pagination exit — introduce a reachedCutoff flag that breaks both the inner and outer loops on the first run older than the cutoff, cutting up to 9 unnecessary API calls per guardrail invocation:

  // Before: skipped the run but kept paginating
  if (!Number.isFinite(createdAtMs) || createdAtMs < cutoffMs) continue;
  
  // After: stops all further page fetches
  if (!Number.isFinite(createdAtMs) || createdAtMs < cutoffMs) {
    reachedCutoff = true;
    break;
  }

• Remove dead guard — countedRuns.length >= maxInspectableRuns in the artifact-inspection loop was unreachable; candidateRuns is already capped to maxInspectableRuns by the time that loop runs.
• Trim per-page logging — replace runIds: runs.map(...) (up to 100 IDs/page) with firstRunId/lastRunId.
• New test — verifies that a page containing one recent run followed by one stale run causes exactly one listWorkflowRuns call and one getRunAIC call.