Propagate invocation-cap guardrail signal in Repo Assist failure handling

[Copilot]

Repo Assist runs could hit Copilot HTTP 429 while consuming only a small fraction of configured max AI credits, but the workflow wasn’t surfacing the explicit budget/invocation-cap guardrail signal in failure reporting. This made rate-limit incidents appear ambiguous in generated issue diagnostics.

• Failure-signal wiring

  • Added missing env passthrough in conclusion → Handle agent failure so failure reporting receives the explicit guardrail signal:
    • GH_AW_INVOCATION_CAP_EXCEEDED: ${{ needs.agent.outputs.invocation_cap_exceeded || 'false' }}

• Behavioral impact

  • Aligns failure classification with agent outputs when 429 is tied to invocation-cap/budget guardrails.
  • Prevents “rate limit without budget guardrail signal” mismatches in Repo Assist incident issues.

• Patch excerpt

  env:
    GH_AW_AI_CREDITS_RATE_LIMIT_ERROR: ${{ needs.agent.outputs.ai_credits_rate_limit_error || 'false' }}
    GH_AW_MAX_AI_CREDITS: ${{ vars.GH_AW_DEFAULT_MAX_AI_CREDITS || '1000' }}
    GH_AW_INVOCATION_CAP_EXCEEDED: ${{ needs.agent.outputs.invocation_cap_exceeded || 'false' }}

• Fixes [aw] AzRetirementMonitor Repo Assist hit AI credits rate limit #73