Synchronous websocket write

[mvollrath]

Fixes #212 by blocking send_message() while a message is being written.

In practice this connects the backpressure to either the rospy.Subscriber Thread (in the case of MessageHandler or ThrottleMessageHandler) or the QueueMessageHandler Thread in that case.

Applications should subscribe with queue_length of at least 1 to prevent pushing all the way back to the rospy.Subscriber, which is probably rude to other clients, and/or use the "workersocket" transportLibrary to decouple the WebSocket from the client main thread.

There's still a way to DoS other clients (0 queue_length, never ACK) but this is a net positive for the websocket server:

• QueueMessageHandler actually works instead of dumping messages into Tornado's limitless queue immediately.
• Server doesn't OOM when the client can't keep up with incoming data.

[jihoonl]

It looks good. Will this work with tornado 4.2.1(the default tornado version in xenial)?

[mvollrath]

Just tested in a kinetic docker with 4.2.1 👍