Fix responses transformer to properly close reasoning before message content

[kwanLeeFrmVi]

Problem

When models output native reasoning_content (e.g., ollama/glm-5, ollama/kimi-k2.5), the Responses API transformer was not properly transitioning from reasoning to message content. The reasoning stream stayed "open" until finish_reason arrived, causing out-of-order events that confused clients and prevented message rendering after reasoning.

Root Cause

In createResponsesApiTransformStream, closeReasoning() was only called at:

1. </thinking> tag detection (for tag-based reasoning)
2. finish_reason arrival
3. Stream flush

When native reasoning_content was followed by content, there was no explicit transition.

Fix

Added early reasoning closure when content arrives after reasoning_content:

if (delta.content) {
  // Close reasoning if we had reasoning_content and now have content
  if (state.reasoningId && !state.reasoningDone) {
    closeReasoning(controller);
  }
  // ... handle content
}

[kwanLeeFrmVi]

Description for commit #972363

Fix: Add mandatory budget_tokens for Claude thinking config and reasoning_effort support

Problem

Claude-compatible providers return 422 error when thinking is enabled without budget_tokens:

Failed to deserialize the JSON body into the target type: thinking: missing field budget_tokens at line 1 column 111088

Additionally, Claude Code clients sending reasoning_effort (e.g., reasoning_effort: "high") were not having this converted to the appropriate thinking.budget_tokens value.

Solution

open-sse/config/runtimeConfig.js

• Added DEFAULT_BUDGET_TOKENS = 16000 constant for default thinking budget

open-sse/translator/request/openai-to-claude.js

• Import DEFAULT_BUDGET_TOKENS from runtime config
• Fix: Always include budget_tokens in thinking object (was previously omitted when not explicitly provided)
• Feature: Convert reasoning_effort to thinking.budget_tokens:
  • low → 1024 tokens
  • medium → 8192 tokens
  • high → 32768 tokens

Files Changed

• open-sse/config/runtimeConfig.js
• open-sse/translator/request/openai-to-claude.js