Basic StreamExt::{flatten_unordered, flat_map_unordered} impls

[olegnn]

Closes #1781
Closes #2378

Basic implementations of flatten_unordered and flat_map_unordered using FuturesUnordered in combination with custom poll state and wakers under the hood. I didn't measure performance but I think it's fast enough. It differs from flatten and flat_map because it polls stream and inner streams concurrently.

Edited: I added benchmarks, which can be compared with futures_unordered. flatten_unordered with a stream of 10000 oneshots each of which produces an inner stream with one item taking approx about x1.5 time of the original futures_unordered benchmark with 10000 oneshots.

New update: after changing the logic to make a stream being polled in a loop instead of once per poll_next, flatten_unordered takes about 1 - 1.1x time of the original futures_unordered benchmark.

[taiki-e]

Looks great!

[taiki-e]

Thanks!

[taiki-e]

Published in 0.3.20.