Harden APP discussion bot Copilot parse path against token/model failures

[Copilot]

The APP discussion bot intermittently failed in actions/ai-inference@v3 after invoking Copilot CLI (copilot -p ...), with the parse step exiting non-zero. This change reduces hard failures in the inference path while preserving the existing fallback behavior.

• Inference-step gating on credential availability

  • Run Copilot-dependent steps only when COPILOT_PAT is configured.
  • Prevents unnecessary CLI invocation in environments where inference credentials are absent.

• Model compatibility adjustment

  • Switch AI inference model from gpt-5.5 to gpt-5 in the parse step to align with more reliable Copilot CLI execution.

• Behavioral impact

  • When Copilot inference cannot run, the workflow now cleanly follows the existing URL-based parsing/fallback path instead of surfacing avoidable parse-step failures.

- name: Parse comment (GitHub Copilot)
  if: steps.ctx.outputs.skip != 'true' && secrets.COPILOT_PAT != ''
  uses: actions/ai-inference@v3
  with:
    system-prompt-file: .github/prompts/parse-app-submission-system.md
    prompt-file: runtime-parse-prompt.txt
    model: gpt-5