ci(upstream-merge): add chat-completions fallback provider and force it via -c model_provider=openai_chat to avoid Responses API 500s via proxy; keep proxy for both paths

.github/workflows/upstream-merge.yml

@@ -440,6 +440,19 @@ jobs:
           # Wait briefly for readiness
           for i in {1..30}; do if nc -z 127.0.0.1 5055; then break; else sleep 0.2; fi; done || true
 
+      - name: Configure Code to use Chat Completions fallback
+        if: steps.check_upstream.outputs.skip != 'true'
+        run: |
+          set -euo pipefail
+          mkdir -p "$HOME/.codex"
+          cat > "$HOME/.codex/config.toml" <<'TOML'
+          # CI-local provider that forces the classic Chat Completions wire API via the proxy.
+          [model_providers.openai_chat]
+          name = "OpenAI (Chat)"
+          base_url = "http://127.0.0.1:5055/v1"
+          wire_api = "chat"
+          TOML
+
       - name: Print proxy startup log tail
         if: steps.check_upstream.outputs.skip != 'true'
         run: |
@@ -679,6 +692,7 @@ jobs:
             npm_config_cache="$SAFE_HOME/.npm" \
             npx -y @just-every/code@latest \
             exec \
+            -c model_provider=openai_chat \
             -s workspace-write \
             -c sandbox_workspace_write.allow_git_writes=true \
             -c sandbox_workspace_write.network_access=true \