fix(cron): avoid provider summaries for script jobs

[yinkev]

Summary

• skip provider-specific failure classification for no_agent=True cron jobs;
• preserve existing provider rate-limit, timeout, and authentication summaries for agent-backed jobs;
• let script-only failures report their actual collapsed stdout instead of a false provider diagnosis.

Fixes #70908.

Root cause

The failure summarizer scanned the complete script stdout for bare 401, 403, 429, quota, and timeout text without considering execution mode. A script-only job can print those strings in successful checks, then fail for an unrelated reason. Since no_agent jobs never call a model provider, every provider classification on that path is structurally wrong.

Verification

• regression was red on current main for authentication, quota, and timeout text;
• focused summary plus complete no_agent cron suite: 22 passed;
• Ruff, py_compile, and git diff --check: clean.

[alt-glitch]

This was generated by AI during triage.

Duplicate of #70977. Both PRs implement the same no_agent guard around provider-error classification in cron/scheduler.py and cover the same false 401/403/429/timeout summaries; #70977 is the earlier open PR.

[yinkev]

Closing in favor of the earlier #70977 after validating its patch on current main. The production change is equivalent, while #70977 has broader explicit coverage for 401, 403, 429, timeout, agent-backed jobs, and legacy job dictionaries. Its focused current-main suite passes 51/51, so this PR no longer preserves unique value.