Consider optimizations to reduce the number of host calls

While host calls are pretty fast these days, it's still useful to consider optimizations to reduce their number per request for really low-latency scenarios. This has come up in two areas so far:

• @PiotrSikora mentioned in this comment that many requests have empty bodies, so perhaps consume could short-circuit the hop through incoming-stream, maybe with a variant like this.
• @BrendandBurns mentioned in this comment the sequence of calls to create headers, then create a response, then make the request. Perhaps we could have some more one-shot options that do multiple of these at once.

Both seem like potentially good ideas to me but I'd be happy to hear more thoughts.

Start with the linked PR #3 discussions and trace the consume/incoming-stream path, along with the sequence that creates headers, creates a response, and makes the request. Define which host-call reductions are viable for empty bodies and one-shot request setup, then document and test the agreed API changes.