feat(minimax-multimodal-toolkit): update models, simplify scripts, and standardize docs

[BLUE-coconut]

Summary

• Add default models table & error handling: New section documenting default models for each capability (TTS, Music, Image, Video) so the agent uses them without prompting the user. Added auto-fallback logic for video quota exhaustion (MiniMax-Hailuo-2.3 → MiniMax-Hailuo-2.3-Fast).
• Remove deprecated models & simplify constraints: Removed speech-2.6, T2V-01, I2V-01, S2V-01, MiniMax-Hailuo-02 references. Unified all video modes (t2v/i2v/sef/ref) to use MiniMax-Hailuo-2.3 with 6s + 768P only. Changed default video duration from 10s to 6s.
• Translate Chinese text to English: All Chinese descriptions in docs, prompt examples, camera instructions, and aspect ratio tables are now in English for international accessibility.
• Simplify image generation script: Replaced the Windows-oriented temp-file payload builder with inline jq piping. Fixed base64 -w 0 → base64 for macOS compatibility. Removed verbose error output.
• Add model validation to video scripts: Both generate_video.sh and generate_long_video.sh now validate model/duration/resolution combinations upfront with clear error messages.
• Remove plan limits section: Quota tables were removed from SKILL.md as they are subject to frequent changes and add maintenance burden.

Test plan

• Verify TTS generation works with speech-2.8-hd default model
• Verify image generation (t2i and i2i modes) with simplified payload builder
• Verify video generation in all modes (t2v, i2v, sef, ref) defaults to MiniMax-Hailuo-2.3 at 6s + 768P
• Verify model validation rejects unsupported model/duration/resolution combinations
• Verify long video generation defaults to 6s segments
• Confirm base64 command works on macOS without -w 0 flag

🤖 Generated with Claude Code