feat(vlm): add streaming response handling for OpenAI VLM

[KorenKrita]

Description

Some OpenAI-compatible APIs (e.g., certain third-party proxies or gateways) return SSE streaming responses even when stream=False is requested. This causes the existing OpenAIVLM backend to crash with AttributeError when it tries to directly access response.choices[0].message.content.

This change adds automatic detection and adaptation to OpenAIVLM: when the API returns a streaming response regardless of the requested format, the backend now transparently consumes the stream, concatenates the content, and correctly extracts token usage. Normal non-streaming responses are unaffected — no caller changes required.

Related Issue

Type of Change

• Bug fix (non-breaking change that fixes an issue)

• New feature (non-breaking change that adds functionality)

• Breaking change (fix or feature that would cause existing functionality to not work as expected)

• Documentation update

• Refactoring (no functional changes)

• Performance improvement

• Test update

Changes Made

• Add _is_streaming_response() and _is_async_streaming_response() to detect streaming vs non-streaming responses by checking __iter__/__aiter__ and choices attributes, with exclusion of basic types (str/list/dict)
• Add _extract_content_from_chunk() and _extract_usage_from_chunk() helpers to extract content and token usage from individual SSE chunks
• Add _process_streaming_chunks() sync method that consumes all chunks, concatenates content, and records the last non-zero usage
• Add _extract_content_and_usage() / _extract_content_and_usage_async() unified entry points that auto-select streaming or non-streaming processing based on response type
• Add _handle_response() / _handle_response_async() and _finalize_response() for common post-processing (empty response warnings, token usage update)
• Refactor get_completion, get_completion_async, get_vision_completion, get_vision_completion_async to use the new response handling pipeline instead of direct attribute access
• Add tests/unit/test_openai_vlm_streaming.py with 5 test classes and 14 test cases covering:
  • Streaming/non-streaming detection (including __iter__/__aiter__/_iterator/basic type exclusion)
  • Chunk content and usage extraction
  • Sync and async streaming consumption with content concatenation
  • Full Mock OpenAI client integration tests (text completion / vision completion / token usage tracking)

Testing

• I have added tests that prove my fix is effective or that my feature works

• New and existing unit tests pass locally with my changes

• I have tested this on the following platforms:

  • Linux

  • macOS

  • Windows

Checklist

• My code follows the project's coding style

• I have performed a self-review of my code

• I have commented my code, particularly in hard-to-understand areas

• I have made corresponding changes to the documentation

• My changes generate no new warnings

• Any dependent changes have been merged and published

Screenshots (if applicable)

Additional Notes