Socket reconnection on retries of JoinGroup, SyncGroup, LeaveGroup

[saich]

Today, if JoinGroup sees a socket disconnection, it will retry again without reconnecting the connection which results in subsequent retries all to fail.

This change will ensure that we attempt to reconnect the connection if see a disconnected connection while selecting a unique RequestHandler to use for this request.

Alternative approaches that can be considered:

• Reconnect automatically in RequestHandler.request before calling self.shared.requests.put
• Reconnect automatically in BrokerConnection.request if required

[codecov-io]

Codecov Report

Merging #939 into master will increase coverage by 0.09%.
The diff coverage is 66.66%.

@@            Coverage Diff             @@
##           master     #939      +/-   ##
==========================================
+ Coverage   83.32%   83.41%   +0.09%     
==========================================
  Files          36       36              
  Lines        3807     3811       +4     
  Branches      562      563       +1     
==========================================
+ Hits         3172     3179       +7     
+ Misses        489      486       -3     
  Partials      146      146

Impacted Files	Coverage Δ
pykafka/managedbalancedconsumer.py	84.1% <100%> (ø)	⬆️
pykafka/broker.py	88.13% <60%> (-5.51%)	⬇️
pykafka/connection.py	83.16% <0%> (-1.99%)	⬇️
pykafka/topic.py	80.41% <0%> (-1.04%)	⬇️
pykafka/balancedconsumer.py	91.22% <0%> (+1.16%)	⬆️
pykafka/simpleconsumer.py	86.6% <0%> (+1.56%)	⬆️
pykafka/cluster.py	71.89% <0%> (+1.82%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 6599354...238bcc8. Read the comment docs.

[emmettbutler]

Thanks for the contribution. I think your first alternate suggestion of putting this functionality in RequestHandler might be a better way to maintain encapsulation.