Backend test parametrization #271

Now that almost all of HTTPX internals are relying on the concurrency backend, and before tackling the Trio backend (#120), it would be good to parametrize all tests with a backend fixture.

For now, it would only point to AsyncioBackend, but we're going to need it for TrioBackend anyway — the main goal is to reduce clutter when the trio backend lands a PR.

Once the backend fixture is defined, any test that uses AsyncClient, HTTPConnection or ConnectionPool can basically be parametrized with it, e.g.:

# conftest.py
@pytest.fixture(params=[pytest.param(AsyncioBackend, marks=pytest.mark.asyncio)])
def backend(request):
    backend_cls = request.param
    return backend_cls()

# test_async_client.py
async def test_get(backend):
    async with AsyncClient(backend=backend) as client:

I'll tackle this look soon, but if anyone has thoughts on this approach let me know. :-)