feat: add ModelScope as a new LLM provider

[liyidd]

feat: add ModelScope as a new LLM provider

Summary

Adds ModelScope (魔搭社区) as a high-performance LLM provider to Bolt.diy, focusing on the latest Qwen3 series and MoE (Mixture of Experts) architectures optimized for mainland China connectivity and developer efficiency.

Changes

New Provider (1)

• ModelScope - Leading open-source model hub in China, providing low-latency access to state-of-the-art Qwen3 models via an OpenAI-compatible API.

Model Enhancements

Integrated the latest Qwen3 flagship models via ModelScope's infrastructure:

• Qwen3-235B-A22B (MoE): The current SOTA for coding and reasoning, featuring an efficient MoE architecture that balances deep intelligence with responsive inference.
• Qwen3-32B: High-performance balanced model, ideal for rapid prototyping and medium-scale code analysis.
• Qwen3-14B: Ultra-fast coding assistant, optimized for real-time auto-completion and simple refactoring tasks.

Total Models Added

• New Provider: 3 core Qwen3 variants (Expandable via dynamic model discovery).

Coding Model Performance

Model	Performance	Provider
Qwen3-235B-A22B	SOTA Coding & MoE reasoning, 128K+ context	ModelScope
Qwen3-32B	Excellent logic-to-latency ratio	ModelScope
Qwen3-14B	Fastest response for localized dev loops	ModelScope

Technical Details

• Architecture: Implemented ModelScopeProvider using the standard OpenAI-compatible protocol at api.modelscope.cn/v1.
• Environment Variables: Added MODELSCOPE_API_KEY to support authenticated requests.
• Discovery: Supports both a static model list for core Qwen3 models and dynamic model discovery via the /v1/models endpoint.
• Registry: Auto-registered through LLMManager following the project's existing provider pattern.

Test Plan

• ✅ Type checking passed: Verified against internal TS definitions (resolving prior build/server conflicts).
• ✅ Linting passed: Fixed @blitz/lines-around-comment and Prettier indentation errors.
• ✅ Provider Export: Successfully exported from the provider registry.
• ✅ Documentation: Updated .env.example with the new environment variable.

Coding Focus

This PR enhances Bolt.diy's accessibility for developers in regions with restricted global LLM access by adding:

• Optimized Connectivity: Low-latency endpoints via ModelScope's China-based infrastructure.
• Next-Gen Qwen Support: Direct access to Alibaba's latest Qwen3 open-source weights.
• Large Context Handling: Fully supports Qwen3's extended context window for analyzing complex project structures.

Breaking Changes

• None — This is a purely additive change.

Related Issues

• None