Responses minor fix + UTs

[michalkulakowski]

🛠 Summary

JIRA/Issue if applicable.
Describe the changes.

🧪 Checklist

• Unit tests added.
• The documentation updated.
• Change follows security best practices.
  ``

[Copilot]

Pull request overview

This PR updates OVMS’s OpenAI Responses behavior to better match vLLM/OpenAI client expectations around (1) how historical reasoning is replayed in multi-turn conversations and (2) how Responses output[] is serialized when the model emits tool calls without any text. It also expands test coverage for multimodal image parsing and media authorization allowlists.

Changes:

• Add a post-processing step to drop/strip stale reasoning_content from earlier assistant turns before the last “final answer” assistant turn when building chat/completions-shaped messages from a Responses input[].
• Update Responses serialization to omit the message output item when no text content was produced (tool-calls-only outputs).
• Extend unit tests for multimodal images (base64, local filesystem allowlist, domain allowlist) and for Responses input_image.image_url accepted shapes.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
src/llm/apis/openai_responses.cpp	Adds dropHistoricalReasoning() post-processing to the Responses-input builder sinks; skips emitting empty message output items when no text was generated.
src/test/http_openai_handler_test.cpp	Adds extensive parsing/authorization tests for images and multi-turn reasoning-drop behavior; adds Responses-only coverage for input_image.image_url shape variants and negative cases.