fix: tolerate post-completion stream events in ResponseAccumulator

[zklapow]

Summary

The Responses API sends response.rate_limits.updated events after the terminal response.completed event. ResponseAccumulator.accumulate() throws IllegalStateException("Response has already been completed.") on any event received after a terminal event, which crashes callers that pass every streamed event through it.

This replaces the hard check with an early return so post-completion events are silently ignored and the accumulated response remains intact.

Reproduction

curl https://api.openai.com/v1/responses \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "input": "What is the weather in San Francisco?",
    "model": "gpt-4.1",
    "tools": [{
      "name": "get_weather",
      "type": "function",
      "description": "Get weather",
      "parameters": {"type":"object","properties":{"location":{"type":"string"}},"required":["location"]}
    }],
    "stream": true
  }'

The stream ends with:

event: response.completed
data: {"type":"response.completed", ...}

event: response.rate_limits.updated
data: {"type":"response.rate_limits.updated", ...}

Passing these events to ResponseAccumulator throws:

java.lang.IllegalStateException: Response has already been completed.
    at com.openai.helpers.ResponseAccumulator.accumulate(ResponseAccumulator.kt:119)

Test plan

• Updated accumulateAfterCompleted test to assert post-completion events are ignored rather than throwing
• Verified accumulated response remains intact after receiving post-completion events
• All existing ResponseAccumulatorTest tests pass