Add image token compression with trained ML router

Replace legacy memory system with HierarchicalMemory

Add comprehensive evaluation framework for compression accuracy

Add pre-commit hooks for ruff linting and formatting

Integrate DynamicContentDetector into CacheAligner (Phase 1)

Fix HeadroomAgnoModel to extend agno.models.base.Model and add real integration tests

Enhance /stats endpoint and add comprehensive benchmarks

Add TOIN field-level learning and comprehensive integration tests

Fix mypy type errors in LiteLLM and OpenAI providers

Add Agno agent framework integration

Use LiteLLM for model pricing instead of hardcoded values

Add universal compression module with ML-based content detection

Add persistent memory system with zero-latency inline extraction

Add seamless LangChain integration

Fix all ruff lint and format errors for CI