Tokiobeam-channel

This project aims to combine tokio (>=0.2.0) with crossbeam (0.7.3) to provide a faster channel.
current performance statistics looks like this:

Transfering 80,000 i32 numbers, calling after(200ns) per message:

two recv threads:   21787ns/iter, total  1743007247ns
single recv thread: 48083ns/iter, total  3846659745ns

Transfering 200,000 i32 numbers, using unbounded channel

tokiobeam-channel:  81ns/iter, total 16226397ns
tokio-channel:     235ns/iter, total 47148280ns

Transfering 200,000 i32 numbers, using bounded channel(1)

tokiobeam-channel:  340ns/iter, total    69172570ns
tokio-channel:     8641ns/iter, total  1728361297ns

Transfering 200,000 i32 numbers, using bounded channel(1000)

tokiobeam-channel:  53ns/iter, total 10774397ns
tokio-channel:     304ns/iter, total 60842293ns

to run the benchmark, you need to turn on the nocapture to see the messages.
This is because test::Bencher doesn't have good support on async/await that we need to estimate on our own.

cargo bench -- --nocapture --color=always

Features:

• cloneable Receiver/UnboundedReceiver
• faster AtomicWaker (AtomicSerialWaker) that supports FIFO
• supports newest tokio and async/await calls
• supports bounded channel with zero capacity (core thread will be blocked, so make sure there are enough threads to run all Sender/Receivers on them)
• less cpu usage

Please report if you found any race condition issue.