Add MiniMax-M3 to the LLM model dropdown

[octo-patch]

Reason: The executable MiniMax model dropdown listed only the MiniMax-M2.7 entries and omitted MiniMax-M3, so that model could not be picked in the LLM clipping UI.

Changes

• funclip/launch.py: add minimax/MiniMax-M3 to the llm_model dropdown choices, next to the existing MiniMax entries.
• tests/test_minimax_launch_integration.py: update the dropdown regression test so the listed MiniMax models are asserted to include minimax/MiniMax-M3.

No routing changes were required. funclip/llm/openai_api.py already strips the minimax/ prefix and resolves any MiniMax model to the OpenAI-compatible endpoint (https://api.minimax.io/v1 by default, with MINIMAX_API_BASE available for the regional host and MINIMAX_API_KEY for credentials), and llm_inference in funclip/launch.py already dispatches the minimax/ prefix through openai_call. The dropdown entry was the only missing piece.

Checks

• python -m pytest -q tests/test_minimax_launch_integration.py tests/test_minimax_api.py tests/test_openai_api.py — 10 passed.
• python -m pytest -q tests/ --ignore=tests/test_duplicate_text_matching.py --ignore=tests/test_model_selection.py --ignore=tests/test_recognition_result_compat.py — 47 passed, 1 skipped, 6 failed. All 6 failures are ModuleNotFoundError for gradio and twelvelabs, which are not installed in this sandbox; the same 6 tests fail identically on the base commit, so they are unrelated to this change. The three ignored modules fail to import for the same reason (gradio, moviepy).
• python -m compileall funclip/launch.py tests/test_minimax_launch_integration.py