This closes #1109

This closes #1103

..capable of tracking min, max and estimates for count distinct and
heavy hitters.

The count distinct algorithms are Linear Counting and HyperLogLog, both
from an imported library from clearspring.

The heavy hitters algorithms are Lossy counting (Manku et.al 2002) and
one based on Count Min Sketch (Cormode 2005).

The heavy hitters algorithms are implemented in the statistics package
in the flink-operator-stats submodule of flink-contrib.

Include tests verifying if merged skecthes have the same guarantees as
local sketches.

Added conditional to every collect variable, to make stats more
configurable. Made print operator stats more informative. Added
arguments to yarn pom to enable building

Implemented deep cloning for class OperatorStatistics. Added additional
tests in OperatorStatsAccumulatorTest to check if the accumulator works
when only one statistic is being tracked, rather than all.

This closes #605.

This closes #1094

This closes #1091

This also bumps surefire/failsafe version to 2.8.1

This closes #889.

[FLINK-2580] [runtime] Expose more methods form Hadoop output streams and exposes wrapped input and output streams.

This closes #1093

This closes #992.

This closes #1052

- improve test layout

This closes #1073.

[FLINK-2570] [gelly] Added a description of the I/O

This closes #1054

[FLINK-2567] [core] Allow quoted strings in CSV fields to contain quotation character inside of the field, as long as its escaped Ex: 'Hi my name is \'Flink\''

This closes #1059

This closes #1082

This closes #1096

The test is inconclusive when the test failure happens before the first checkpoint.

LocalExecutor and Client now pass Configuration to JobGraphGenerator.

This closes #1095

This closes #1089.

[FLINK-2161] [scala shell] Modify start script to take additional argument (-a <path/to/class> or --addclasspath <path/to/class>) for external libraries

This closes #805

This closes #1074

- call new start() method of FlinkMiniCluster

This closes #1085.

[hotfix] [gelly] Remove unintended calls to getExecutionEnvironment in VertexCentricIteration and GatherSumApplyIteration

This closes #1031

[FLINK-2448] [tests] Create new Test Environments on every call to getExecutionEnvironment and allow sharing of job execution results

This closes #1067

This prevents possible null pointers when the JVM shuts down before the I/O manager was fully started.

* modified algorithm as explained in the issue
* updated method documentation

[FLINK-2590] reducing required bit shift size

* maximum bit size is changed to getNumberOfParallelSubTasks() - 1

This closes #1075.

This closes #1076