trio.timeit utility

Trio should offer an async timeit utility. It's useful for measuring performance of "fast" async calls such as sleep(0).

It's important for both trio devs and users to understand the overhead of checkpoints. E.g. for users: in situations where you're quickly draining an async queue, or when you're deciding how frequently to call sleep(0) in some CPU bound loop.

Here's an example using the pytest harness. Ideally we'd have a module with the same API as timeit, only running within a trio async context.

async def test_sleep0_perf():
    n = 10
    dt = 0
    while True:
        t0 = perf_counter()
        for _ in range(n):
            await trio.sleep(0)
        dt = perf_counter() - t0
        if dt < 0.2:
            n *= 10
        else:
            break
    print(f'{1e6 * dt / n: .1f} usec average in {n} calls')
    assert False

Start by comparing Python's timeit API with the pytest-harness example in the issue, especially the perf_counter loop around trio.sleep(0). Check how an async utility would run inside a Trio context and what API compatibility is required. Done means the proposed utility measures fast async calls and has coverage for the sleep(0) use case.