WHAT:
- Support customizable LLM selection for each processing stage
- Optimize model selection based on task requirements

WHY:
- Enable cost optimization by using appropriate models for different tasks
- Improve translation quality with high-end models where needed

DETAILS:
- Recommend high-quality models (e.g. claude-3-5-sonnet) for translation tasks
- Suggest cost-effective models (e.g. Yi-Lightning) with JSON output support for simple operations
- Refactor config structure to support per-stage LLM configuration
- Update core modules to respect stage-specific LLM settings

AFFECTED:
- config.yaml: Add per-stage LLM configuration
- core/: Update LLM handling in processing modules
- st_components/: Update UI for LLM selection

Breaking Changes:
- Configuration file structure updated for LLM settings
- LLM initialization process modified