historical data restriction 3 calls per second

If you need to call that many instruments with the throughput that you are expecting, I am assuming you are using it for live(actual trading) or forward testing. In that case, historical data is not the way to go. When your application comes up, lets say 9:00 AM in the morning, let it query historical for all the 1000 instruments giving proper time gap using any standard design pattern like leaky bucket. This would ensure by 9:15 you have stock's previous candles. Now when 9:15 starts, we would need to start the websocket and allow it to create the candles on the fly from the tick data and append this candle data to the previously got data from the historical candles and let it run like this for the course of the day. That way, you get best of the both.