Refactor how HTTP body can be consumed as raw stream

We have some API endpoints that download HTTP bodies in raw format. There, we usually copy the body before returning allocating memory. The idea is to check what we can do to reduce this.