Overhead of creating new threads in Streaming Response

**Is your feature request related to a problem? Please describe.**
There is an overhead of creating new threads when using streaming response feature.
This drogon example demonstrates it very well: https://github.com/drogonframework/drogon/blob/8541e671430f9605a62f1afbd786c1dbd54197d6/examples/async_stream/main.cc#L15-L24

We have uncontrolled number of new threads being created which is a bottleneck for high-throughput workloads.
Another issue is that the thread is detached and never joined.

**Describe the solution you'd like**
Other http frameworks use internal event loop to schedule partial send operation, example:
https://github.com/tensorflow/serving/blob/8eec641dee627243e13f8d04d8980f326c3e371d/tensorflow_serving/util/net_http/server/internal/evhttp_request.cc#L359-L360

**Describe alternatives you've considered**
No alternative

**Additional context**
High-throughput scenarios, for example serving LLM text generation with streaming output approach