Merge pull request #641 from mateiz/spark-1124-master

SPARK-1124: Fix infinite retries of reduce stage when a map stage failed

In the previous code, if you had a failing map stage and then tried to run reduce stages on it repeatedly, the first reduce stage would fail correctly, but the later ones would mistakenly believe that all map outputs are available and start failing infinitely with fetch failures from "null". See https://spark-project.atlassian.net/browse/SPARK-1124 for an example.

This PR also cleans up code style slightly where there was a variable named "s" and some weird map manipulation.