Make ASGI dispatcher asyncio-agnostic

Note: this issue breaks apart steps originally taken in #217.

The ASGI dispatcher currently relies on asyncio-specific APIs to perform various operations, which is a blocker towards supporting alternative concurrency backends.

We want to make sure all of the I/O inside ASGIDispatch goes through the backend instead.

This will need the following, most of which can be tackled in independant PRs:

• #257 Hard-code self.backend = AsyncioBackend() on ASGIDispatch to be able to iteratively defer asyncio-specific APIs to the backend.

• #257 Introduce a queue-like interface:

class BaseQueue:
    async def get(self): ...
    async def put(self, value: typing.Any): ...

along with a .create_queue() method on ConcurrencyBackend, to be used instead of/within the BodyIterator.

(For trio we’ll probably need to add a .close() method as well, but that can be confirmed and tackled in due time.)

• #260 Introduce a BaseEvent interface:

class BaseEvent:
    def set(self): ...
    def is_set(self): ...
    async def wait(self): ...

along with a .create_event() method on ConcurrencyBackend to abstract asyncio.Event().

• #261 Remove usage of loop.create_task() via the background manager interface.

• Lastly, turn the hard-coded backend into a parameter. Make sure it is optional and we use AsyncioBackend as a default.

[tomchristie]

Yeah it’ll tie in. We’ll want to get this side of things all nice and clean and sorted first, then we’ll be able to take a proper look at #52 again.

[tomchristie]

Going to see if I can take a crack at dropping the asyncio.FIRST_COMPLETED bit of implementation, since that’d make this issue somewhat narrower in scope.