Fix unclosed generator when running on trio

[florimondmanca]

While taking a look at resolving #2186, I came across an issue in the trio variant of test_asgi_streaming.

It seems that our usage of generators for __aiter__ doesn't satisfy trio's strict semantics.

AFAIU, when __aiter__() does more than delegating to an underlying async iterable, we should be implementing __anext__() so that aclosing() is applied to that streaming. Though my understanding is a bit limited so far in the night. 😃

This comes with a companion PR on httpcore. encode/httpcore#657

[tomchristie]

It seems that our usage of generators for __aiter__ doesn't satisfy trio's strict semantics.

How does this present itself?
Can you demonstrate a warning that we're seeing with the existing implementation?

[tomchristie]

I'm closing this off in line with encode/httpcore#657 for now. I'm open to the suggestion that there may be an issue here, but we'd need to be able to clearly demonstrate that first.