increase backend conversation speed

[dail45]

1. using orjson instead of standart python json, another small changes for better performance
2. accumulating chunks from openai when js in decoding chunks (config->backend_config->join_chunks turn on by default)