Improve tracking failure handling (#1063)

[beckjake]

Fixes #1063

Add a 1s timeout to tracking http calls
Add an on-failure handler that disables tracking after the first failure

To test this, I added this to my /etc/hosts: 127.0.0.1 fishtownanalytics.sinter-collect.com

Then I ran this from a root python prompt:

import socket
s = socket.socket()
s.bind(('localhost', 443))
conns = []
while True:
  conns.append(s.accept()[0])

And ran dbt run with tracking enabled in my profile. Before this change, dbt hangs forever. After, it pauses for 1 second once and then disables tracking until the next dbt invocation.

I chose this model over a thread-based one because it makes process teardown and cancellation a whole lot easier.

The only actual change to http_get in the overridden TimeoutEmitter is the timeout=1.0 argument to requests.get, the rest is just pep8 formatting.

I experimented with various times but I found anything longer than a second felt pretty annoying.

[drewbanin]

One tiny config change, otherwise LGTM! Glad we fixed this one