fix: LLM provider with exponential backoff, streaming, token tracking

fix: Real mutation execution + AST-based mutations + deeper security scanner

fix: Semantic coverage mapping + real mutation execution + syntax validation

Add services __init__.py"

MAJOR: Add multi-agent iterative refinement engine (MuTAP-inspired) with behavior coverage mapping

Add 5 novelty features: complexity analyzer, gap detector, quality scorer, mutation testing, security scanner

Add backend/app/services/llm_provider.py — universal LLM provider + complete API

Add prompt builder — expert QA engineer system prompts for each input type

Add test generator service — core AI test generation logic

Add core services: LLM provider, test generator, API analyzer, prompt builder