feat: add MiniMax as first-class LLM provider (M3 default)

[octo-patch]

Summary

Upgrade MiniMax provider to use M3 as the default model. M3 is MiniMax's latest flagship LLM with a 512K context window, 128K max output, and image input support.

This PR also introduces MiniMax as a first-class LLM provider in valuecell (backend factory + YAML config + frontend icon + i18n + README links).

Changes

M3 upgrade (this iteration)

• Add MiniMax-M3 to python/configs/providers/minimax.yaml and set as default_model
• Keep MiniMax-M2.7 and MiniMax-M2.7-highspeed as alternatives
• Remove older MiniMax-M2.5 and MiniMax-M2.5-highspeed models
• Update MiniMaxProvider docstring in factory.py accordingly
• Update unit tests to assert M3 is the default and listed first
• Deduplicate tests/__init__.py (was a copy of test_minimax_provider.py)

Provider integration (initial work, unchanged)

• Backend: MiniMaxProvider class using agno's OpenAILike with temperature clamped to (0.0, 1.0]; registered in ModelFactory._providers
• Config: MiniMax YAML config + config.yaml registration with MINIMAX_API_KEY env var
• Backend utils: JSON mode detection for minimax.io base_url in utils/model.py; API key URL mapping in models router
• Frontend: MiniMax provider icon + icon registration + i18n strings (en, zh_CN, zh_TW, ja)
• Docs: README (EN/ZH/ZH_Hant/JA) provider list updated + CONFIGURATION_GUIDE.md provider table

Model spec (M3)

Field	Value
Context window	512K
Max output	128K
Image input	Yes (image only, no video/audio/documents)
Pricing (per M tokens)	input $0.6 / output $2.4 / cache read $0.12

Why

MiniMax-M3 is the latest flagship model. Defaulting to M3 gives users the best capabilities by default while M2.7 / M2.7-highspeed remain available as alternatives for cost or latency trade-offs.

Testing

• pytest valuecell/adapters/models/tests/test_minimax_provider.py — 35 passed
• Unit tests cover: model creation for M3 / M2.7 / M2.7-highspeed, temperature clamping, YAML default_model == M3, M3 listed first, M3 has 512K context, older models removed.