chore: update model provider presets

[c121914yu]

Summary

• Refresh the registered model provider presets from the latest upstream/main.
• Add gemini-3.6-flash and gemini-3.5-flash-lite, remove seven unavailable Kimi/Moonshot IDs, and correct Groq qwen/qwen3.6-27b output capacity.
• Remove 23 Qwen IDs scheduled for shutdown or already offline by Alibaba Cloud Model Studio.
• Remove four ERNIE IDs marked retired by Baidu Qianfan and update the multimodal capability test for the remaining Qwen visual models.
• Align Baichuan4, Baichuan4-Turbo, Baichuan-M3, and Baichuan-M3-Plus with the official 32k context limit.
• Align Jina reranker v3 with the official 131,072-token input limit.

Verification

• bun run test passed: 40 test files and 249 tests.
• Targeted ESLint passed for the changed provider files.
• Model inventory and duplicate-ID checks passed: 245 entries across 34 registered providers.
• The model-provider plan dry-run and write completed successfully.
• git diff --check passed.
• bun tsc --noEmit remains blocked by the existing sdk/factory/src/tool-factory.test.ts:285 type error; that file is unchanged in this PR.

Official sources

• Baichuan model pricing and context lengths
• Jina embeddings and reranker models
• Jina reranker models
• Alibaba Cloud Model Studio models
• Alibaba Cloud Model Studio model decommissioning
• Baidu Qianfan model retirement record
• Baidu Qianfan current model list
• OpenAI deprecations
• Anthropic model deprecations
• Google Gemini deprecations
• DeepSeek pricing and model docs
• Groq supported models