Explore returning json response instead of pydantic models for better serialization performance

### Description While looking into fastapi discussions I noticed that returning models could result in revalidation of the model though the validation was done during model construction. There were also issues with serialization using jsonable_encoder mentioned in the thread. To validate this I did some combinations benchmark to return pydantic model, serialized response using json and serialized response using orjson after validation for a simple "ui/dags" endpoint. Returning json or orjson encoded response seems to result in 2x requests per second. Tradeoffs * Using orjson results in extra dependency and in that case FastAPI JSONResponse can be used as builtin option. * Skipping validation for response could be * Need to check how automatic API generation works when return value is not specified in method. Start below as app.py with 10 workers in uvicorn using `uvicorn app:app --workers 10 --port 8000` command. Start a benchmark session for the endpoints with fixed response. Benchmark for different endpoints 1. Return pydantic model with no return value signature 2. Return orjson response with pydantic model in return value signature but no response validation 3. Return orjson response with no model in return value signature but response validated using pydantic 4. Response dict validated with pydantic and dict returned with model in return value signature. 5. Return json response with model validation for response but no model in return signature https://github.com/fastapi/fastapi/issues/1359 https://github.com/fastapi/fastapi/issues/1224#issuecomment-617243856 ``` seq 1 5 | xargs -I{} time hey -c 100 -n 10000 http://localhost:8000/dags\{\} | grep -iE 'Average:|Requests/sec' Average: 0.0137 secs Requests/sec: 5766.3920 1.05user 0.47system 0:01.75elapsed 86%CPU (0avgtext+0avgdata 20088maxresident)k 0inputs+0outputs (1major+4194minor)pagefaults 0swaps Average: 0.0059 secs Requests/sec: 15063.7350 0.86user 0.34system 0:00.68elapsed 178%CPU (0avgtext+0avgda…