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.
• Add Anthropic claude-opus-5 with the official 1M context and 128k maximum output limits.
• Add MiniMax M2-her with the official 64k context and 2,048-token completion limit.
• Add Qwen qwen3.7-flash with the official 1M context and 64k maximum output limits.
• Add OpenAI gpt-5.3-codex with the official 400k context and 128k maximum output limits for agentic coding workflows.
• Audit on 2026-07-29 confirmed no additional in-scope additions or removals; the current registry remains at 249 entries across 34 registered providers.
• Sync on 2026-07-30 merged the latest upstream/main at 8885bcc into this PR branch before the audit.
• Audit on 2026-07-30 confirmed no additional in-scope additions or removals; the registry remains at 249 entries across 34 registered providers.
• Sync on 2026-07-31 merged the latest upstream/main at fc1a1aa into this PR branch before the audit.
• Audit on 2026-07-31 confirmed no additional in-scope additions or removals; OpenAI gpt-5.6 was retained because the official docs define it as an alias for gpt-5.6-sol.

Verification

• bun run test passed: 40 test files and 250 tests.
• Targeted ESLint passed for the changed provider files in the model-update commits.
• Model inventory and duplicate-ID checks passed: 249 entries across 34 registered providers.
• The 2026-07-31 model-provider plan dry-run and write completed with no file changes; 32 provider sources were checked and Moka/Other remain pending without source hints.
• git diff --check passed.
• bun tsc --noEmit remains blocked by the existing sdk/factory/src/tool-factory.test.ts:285 type error after syncing latest main; that upstream baseline issue is unrelated to the model provider changes.

Official sources

• OpenAI model catalog
• OpenAI GPT-5.6 Sol model
• OpenAI model guidance
• Alibaba Cloud Model Studio models
• Alibaba Cloud Model Studio model decommissioning
• Anthropic Claude models overview
• Google Gemini models
• Google Gemini deprecations
• MiniMax text-chat models
• Kimi model list
• Mistral models overview
• Baichuan model pricing and context lengths
• Jina embeddings and reranker models
• Jina reranker models
• Baidu Qianfan model retirement record
• Baidu Qianfan current model list
• DeepSeek model list
• Groq supported models