feat: track cap-exceeded failures via centralized rollup issue, raise cap to 50/24h

[Copilot]

When a failure category hits the 24h issue cap, the conclusion job silently warned and dropped the event. These failures were invisible unless you caught the workflow log. The cap was also set too low at 25.

Changes

handle_agent_failure.cjs

• Raise FAILURE_ISSUE_CATEGORY_DAILY_CAP 25 → 50
• Add DAILY_CAP_ROLLUP_TITLE / DAILY_CAP_ROLLUP_LABEL constants for the new centralized issue
• Add findOrCreateDailyCapRollupIssue(owner, repo) — finds or creates a single open [aw] Daily failure issue cap exceeded issue (non-fatal; errors logged as warnings)
• Replace the warning-only early return with: emit warning → find/create rollup issue → post a comment containing workflow name, run URL, and capped category counts

// Before: silently dropped
core.warning(`Daily per-category issue cap reached for ${summary}.`);
return;

// After: still tracked in a centralized issue
const rollupIssue = await findOrCreateDailyCapRollupIssue(owner, repo);
if (rollupIssue) {
  await github.rest.issues.createComment({ issue_number: rollupIssue.number, body: commentBody });
}
return;

handle_agent_failure.test.cjs

• Update existing cap test: asserts rollup issue is created and comment is posted to it
• Add test for rollup-reuse path: existing rollup found → no new issue created, comment posted to existing