feat: unified prompt with state visibility

fix: logic for ros array mapping and loop guard progress evaluation

feat: unified prompt with state visibility

fix: extend loop guard to cover ROS-stage repeats

refactor: remove regex extraction, pure LLM-driven

feat: stage-specific prompts + contextual ROS

fix: frontend retry + ROS forcing + test update

fix: force hardcoded ROS questions + update test

fix: ROS hallucination guard + debug logging

fix: debug logging + real error messages + ROS guidance

fix: ROS guidance + done-state guard in API

fix: field descriptions + enhanced loop guard

fix: three-layer defense against infinite loop

fix: added new code

fix: use /api/chat endpoint

fix: docker issue

feat: migrate inference engine to Ollama for 10x faster CPU inference

fix: emergency issue

chore: add more lgging

fix: optimize loading

feat : add dual agent architecture

fix: imporve responses flow

fix: add interrupts in graph

fix: imoprve llm propmts

feat: add chat interface

feat: add FastAPI app and CLI entry point

feat: add LLM providers and graph orchestration

feat: add pydantic schemas and state definitions