fix(ai): preserve terminal reasoning metadata - #44997

Summary

• reconcile pending reasoning items against authoritative response.completed.response.output when providers omit response.output_item.done
• preserve terminal encrypted reasoning metadata in canonical reasoning events, assistant messages, and subsequent stateless requests
• avoid duplicating reasoning items already finalized normally and preserve completed-output ordering alongside pending function calls
• cover native OpenAI Responses and generic Responses-compatible provider metadata namespaces

Before and after

Some Responses streams start a reasoning item and stream its summary but omit the individual item-completion event:

{ "type": "response.output_item.added", "item": { "type": "reasoning", "id": "rs_1", "encrypted_content": null } }
{ "type": "response.reasoning_summary_text.delta", "item_id": "rs_1", "delta": "Checked the diff." }
{ "type": "response.completed", "response": { "output": [
  { "type": "reasoning", "id": "rs_1", "encrypted_content": "terminal-state", "summary": [{ "type": "summary_text", "text": "Checked the diff." }] }
] } }

Before: Generic stream finalization closed the reasoning block without its authoritative terminal metadata, so the assistant message retained reasoningEncryptedContent: null. A later store: false request omitted the reasoning item because it lacked encrypted continuation state.

After: The terminal output item closes the pending reasoning block with its authoritative metadata:

{ "type": "reasoning-end", "id": "rs_1:0", "providerMetadata": { "openai": { "itemId": "rs_1", "reasoningEncryptedContent": "terminal-state" } } }

The next stateless request can therefore replay the reasoning item:

{ "type": "reasoning", "id": "rs_1", "summary": [{ "type": "summary_text", "text": "Checked the diff." }], "encrypted_content": "terminal-state" }

Already-finalized reasoning items are not emitted twice.

Verification

• bun test test/partial-json.test.ts test/tool-stream.test.ts test/provider/openai-responses.test.ts test/provider/openai-compatible-responses.test.ts test/provider/openai-responses-websocket.recorded.test.ts test/provider/openai-chat.test.ts test/provider/anthropic-messages.test.ts test/provider/bedrock-converse.test.ts test/provider/xai-responses.test.ts test/response.test.ts test/tool-runtime.test.ts --timeout 30000 --only-failures (352 passing)
• RECORDED_PREFIX=openai-responses bun test test/provider/golden.recorded.test.ts --timeout 30000 --only-failures
• bun typecheck from packages/ai
• bunx prettier --check packages/ai/src/protocols/open-responses.ts packages/ai/test/provider/openai-responses.test.ts packages/ai/test/provider/openai-compatible-responses.test.ts