add HTTP API server

• use async architecture (return task id, run task on separate worker), Run 2 docker compose services: 1 is lightweight fastapi server, and second is worker with heavy load
• v1_upd: use gitlab api to get number of running pipelines instead of using redis project.pipelines.list(status='running')
• v1: store state in redis but use python semaphore, but sync it's state with redis to allow restarts
• v1: rate limiter without proper handling of race conditions, just check some key in redis
• v2: handle race conditions
• put every request to queue (always return token, don't return max jobs reached) and then have some worker that reads from that queue and process tasks