fix(api): detect GPT-5 multimodal models

[panDing19]

Summary

• Recognize newer OpenAI model names when checking multimodal capability, including gpt-4.1 and the gpt-5 family.
• Add regression coverage for gpt-5, gpt-5.4, gpt-5.5, provider-prefixed openai/gpt-5.5, and the existing gpt-5.5 xhigh style input.
• Update the Unreleased changelog entry for the image-input behavior fix.

Validation

• uv run ruff check src tests scripts
• uv run pytest tests/test_tools/test_image_to_text_tool.py -q
• env -u ANTHROPIC_AUTH_TOKEN -u OPENHARNESS_BASE_URL -u ANTHROPIC_BASE_URL -u OPENAI_BASE_URL -u OPENHARNESS_MODEL -u ANTHROPIC_MODEL -u ANTHROPIC_API_KEY -u OPENAI_API_KEY uv run pytest -q (1160 passed, 6 skipped)
• cd frontend/terminal && npx tsc --noEmit not run; frontend not touched.

Notes

• Related issue: none found.
• Follow-up work: none.