Ensure workers can send data popped off the queue at shutdown

[onlynone]

Previously, any data still in the queue would be handled by the main
thread. But, any data that a worker had already popped off the queue,
but not yet sent on the wire, would be lost.

This change pushes down the logic for stopping to the writer and
worker which now tell the threads to exit and wait for them to do so.

[onlynone]

My last attempt to address this issue in #199 only got part way there. That solved the problem of data being left on the queue not being written, but it didn't solve the issue of data that workers had already popped off the queue and not yet sent at the time the main thread shuts down. That data was still being dropped on the floor.

[onlynone]

The builds that failed, failed with:

InfluxDB::Client
  is expected to be a kind of InfluxDB::Writer::Async
  #write_point
    sends writes to client
    when precision, retention_policy and database are given
      writes aggregate payload to the client (FAILED - 1)
      when different precision, retention_policy and database are given
        writes separated payloads for each {precision, retention_policy, database} set

It seems to me like with an async client, it's possible for two different workers to pick up the two datapoints written to the queue, and they wouldn't be able to aggregate them. Should this really be a test? Or maybe we could keep the test, but set the number of workers to 1?

The tests succeed on my laptop.

[onlynone]

@dmke Have you had a chance to look at this?

[Esity]

@onlynone sorry for the delay on this one. Will take a look at this when I get a chance

[dmke]

LGTM!

[dmke]

Merged in 8df479a. Thank you!