reducing allocations when checking python coroutines

If I do:
```julia
while !Bool(fut.done())
sleep(0.01)
end
```
where `fut` is a future created like:
```julia
aio = pyimport("asyncio")
coro = aio.sleep(1)
loop = aio.Runner().get_loop()
aio.run_coroutine_threadsafe(coro, loop)
```
It allocates linearly the more you sleep.
I tried using a global julia vector `[true]` and updating it in place when the future is done, but it doesn't reduce allocations that much, is there a better way?

No repository file or test is named. Reproduce the polling example using asyncio Future.done(), Runner().get_loop(), run_coroutine_threadsafe(), and Julia sleep(), then measure allocations. Done means finding a supported approach whose allocations do not grow linearly with repeated sleeps and documenting or testing that behavior.

索引モデルが issue の本文から書いたものです。