Preserve MiniMax image and video inputs

[octo-patch]

Reason: MiniMax multimodal requests currently lose image data and reject video content before forwarding.

• Accept video content blocks in message validation.
• Convert image and video sources into URL content parts instead of text placeholders.
• Preserve validated multimodal lists through request normalization.

Checks:

• uv run python -m unittest discover -p 'test*.py' -v
• uv run python -m compileall -q server.py test_multimodal.py

[octo-patch]

Updated the multimodal conversion to preserve documented image and video source options (detail, fps, and max_long_side_pixel) and pushed a new commit. I ran uv run python -m unittest discover -p 'test*.py' -v, uv run python -m compileall -q server.py test_multimodal.py, and git diff --check.