A few cheap wins in the async request path

[WH-2099]

I was profiling the asyncio request path while comparing a couple of HTTP clients.

There are three small bits of work we can avoid without changing the API:

• Do not build a cancellation shield when the pool has no connections to close.
• Do not copy client headers when a request does not add any.
• Pass unencoded response chunks through without the no-op chunking work.

I have these split into three small commits, with regression tests around the behavior-sensitive paths.

Local results from the combined patch:

Scenario	Requests/s	Client CPU/request
c1 / 1 KiB	4,710 -> 4,994 (+6.0%)	210.0 -> 198.1 us (-5.7%)
c32 / 1 KiB	4,568 -> 4,813 (+5.4%)	217.2 -> 206.2 us (-5.1%)
c32 / 64 KiB	3,941 -> 4,114 (+4.4%)	251.7 -> 241.1 us (-4.2%)

This was plain HTTP/1.1 on loopback, CPython 3.14.7, stdlib asyncio, pinned server/client CPUs, and five interleaved 2.5-second runs.

These are hot-path numbers, not a claim about TLS, HTTP/2, or WAN performance.

Seems worth a small PR?