Fix transformers response and improve functionality

[itsmattius]

The transformers service provider is not working correctly and needs significant improvements to match the functionality of other providers.

Current State

• Transformers provider exists but has issues
• Response generation is not working properly
• Inconsistent behavior compared to OpenAI provider

Problems Identified

• Response parsing issues
• Model loading problems
• Memory management issues
• Inconsistent output format
• Performance problems

Desired Improvements

• Fix response parsing and formatting
• Improve model loading and caching
• Optimize memory usage
• Ensure consistent output format
• Add proper error handling
• Implement better model selection

Technical Details

• File: src/services/providers/transformers_service_provider.py:29
• Uses Hugging Face transformers library
• Needs GPU/CPU optimization
• Memory management for large models

Acceptance Criteria

• Fix response parsing issues
• Implement proper model loading
• Add memory optimization
• Ensure consistent output format
• Add comprehensive error handling
• Test with different model sizes
• Benchmark performance improvements