Update safe-output health failure messaging to AI Credits

[Copilot]

Safe Output Health Monitor issues were still explaining proxy budget failures in deprecated effective-token terms. This updates the legacy failure path to use the AI Credits narrative and point to current documentation.

• Failure narrative

  • Reword the legacy effective_tokens_rate_limit_error template to explain the condition as a legacy ET signal while directing users to AI Credits as the primary cost model.
  • Replace “Effective Token Budget Exhausted” messaging with migration guidance toward max-ai-credits.

• Doc links

  • Swap deprecated ET-oriented links for current AI Credits references:
    • AI Credits specification
    • Cost management guide

• Renderer wiring

  • Update handle_agent_failure.cjs template variables to pass the new AI Credits and cost-management links into the rendered issue/comment body.

• Focused coverage

  • Update the existing handle_agent_failure tests to assert the new wording, links, and max-ai-credits guidance.

**⚠️ AI Credits Budget Guidance**: The run hit a legacy effective-token rate-limit signal from the API proxy. gh-aw now uses AI Credits (AIC) as the primary cost metric, so migrate per-run budgeting to `max-ai-credits`.

- Learn about [AI Credits](...)
- `max-effective-tokens` is deprecated; use `max-ai-credits` in workflow frontmatter.
- To budget and optimize this workflow, follow the [cost management guidance](...).