Use blocking call with timeout for republish

[jfontan]

After each retry it waited a fixed amount of time. This caused it to retry at least once per message, that is, a 50 ms extra sleep per job.

Now it blocks for a configurable time if no messages are arriving. No retry is needed and jobs are processes as fast as they arrive.

Previously it processed 20 jobs per second. Now it processes 1220 jobs per second.

Needs #4
Fixes #5

[kuba--]

maybe we can make this function more generic (if it makes sense), e.g.:

func (i *JobIter) nextWithTimeout(timeout time.Duration, fn func(*queue.Job) error) error { } 

[jfontan]

nextWithTimeout is a version of Next (https://github.com/src-d/go-queue/blob/master/amqp/amqp.go#L492-L500) that does not block forever. It's only used in republish buried function so I don't see the benefit of adding a function parameter.

I suppose the best thing to do is to change Next to be able to timeout instead of adding a private method. Maybe adding the timeout value when constructing JobIter. It's not done like this as I didn't want to do bigger changes to the Queue library.