Optimize loop queue of workers

Optimize the logic when putting worker back to sync.Pool

Add doc.go

Refine the logic of sync.Pool

Replace mutex with spin-lock

Update READMEs

worker: decRunning when revertWorker is false

Signed-off-by: Cholerae Hu <[email protected]>

Remove some useless code

A little bit of optimization in periodicallyPurge()

Fixed panjf2000#45 and Releasing v2 module