Don't bother looking at the diff, this is almost a complete rewrite of
the previous Scala API. This uses all the work put into the Java API,
such as TypeInformation, the serializers and comparators and the
operators.

The Scala DataSet and ExecutionEnvironment wrap their respective Java
equivalents. TypeInformation is generated by a macro that uses
TypeInformationGen and other macro support classes. The Java
TypeExtractor is completely bypassed but the TypeInformation and
sub-classes are created by the Scala type analyzer. There is special
support for Scala Tuples in the form of ScalaTupleTypeInfo,
ScalaTupleSerializer, and ScalaTupleComparator.

This also adds tests to flink-scala that are ports of the tests in
flink-java.

There are not yet any Scala specific tests in flink-tests. All the
scala example ITCases are commented out, as well as the examples
themselves. Those will be uncommented once the examples are ported. This
will happen in separate commits.

They were in api.java before but they can be used from Scala,
just like the regular functions.

Previously both sides were always wrapped/unwrapped, even if one side
was a tuple type. This was not compatible with the Scala tuples and is
also not necessary anymore because of the GenericPairComparator.

This is necessary because the previous test didn't work for Scala tuples.

Was TupleComparator before which does not work when used from the Scala API.

I'm doing this in preparation for the Scala API rework, since then I can just copy over
the operators package and use them for Scala.

Before, FlatMapOperator, GroupReduceOperator, MapOperator, and MapPartitionOperator
performed the Type extraction themselves while the other Operators had TypeInformation
parameters. Now the are all unified, which makes it possible to use them from the
Scala API.

Also Key extraction for selector functions is moved outside of Keys.java

Now we have TupleTypeInfoBase, TupleSerializerBase, and TupleComparatorBase. They
are now super classes of TupleTypeInfo and the others.

Also rename compare on DataInputView to compareSerialized because Scala
cannot distinguish between the to compare methods for some reason.

This change is necessary for allowing the Scala API to reuse most of the
functionality.

The GenericPairComparator uses the new extractKeys method of
TypeComparator to compare values of any type. This replaces
TuplePairComparator and some other special-case pair comparators. This
is preparatory work for enabling support for Scala Tuples and POJO
comparators.

This closes #94

This closes #117
* each key is associated with the number of times it was inserted into the accumulator
* backed histogram with a tree map to present the entries sorted by key

This closes #120

This closes #124

Temporarily disabled streaming tests that validate jobgraphs that used to be illegal, but are now legal.

Small fix in integration test

Tolerate concurrent scheduling attempts

Make Co-Location constraints resilient aginst out of order scheduling and depply integrate them with slot sharing
Fix miscellaneous checkstyle errors/warnings

Improved distribution of IDs. Previous implementation lost bits due to double-to-long multiplication and rounding.

Fix comparisons (null pointer safe) in JobManagerITCase

Add proper locality of scheduling tracking to scheduler. Add local scheduling to slot sharing groups.

Add tests for job event classes