Add maximum retries to polling after network error.

[voliva]

A possible solution to #440

I need that, as that infinite loop is giving a hard time to the load balancer when we have a couple of users having this issue.

What it does is that for every connection request, it stops retrying after 3 unsuccessful requests. It's the consumer's responsibility then to decide what to do (e.g. retry again?)

I'm not fully aware of the internals of this library, so I'm not too sure if this is fully safe to merge in, but it seems to cover the basic use case.

[brycekahle]

This be would be a major breaking change for most users. Do you know what network conditions are causing a couple users to continuously get a network error?

[voliva]

I created a gist here, trying to reproduce the issue: https://gist.github.com/voliva/21191f5d1cb1f712a259f81e1b6bfce5

This code may look very fictional though - I just tried to show an oversimplified case. In our solution, the SockJS server is written in Java Spring, and we have a check in a HandshakeInterceptor that may drop the requests if they don't pass a validation case.

Currently, it does abort the request after a while (based on connection RTO timeout), but until it doesn't time out it's constantly creating new requests, which add up to thousands. This PR tries to solve this by also aborting the request after Polling.NETWORK_RETRIES consecutive unsuccessful retries.