- Fix for paths with blank characters
- Fix for single entry wildcard classpaths (http://bugs.java.com, bugID: 7146578)

Remove dup code by delegating call from RemoteExecutor#executePlan to RemoteExecutor#executePlanWithJars.

Author: Henry Saputra <henry.saputra@gmail.com>

Closes #307 from hsaputra/reduce_dup_code_in_executePlan and squashes the following commits:

a98f3b6 [Henry Saputra] Remove dup code by delegating call from RemoteExecutor#executePlan to RemoteExecutor#executePlanWithJars.

This would print a warning if a type could not be analyzed by the Scala
TypeAnalyzer. In such a case analysis is done using the Java
TypeExtractor but the warning seems to confuse people.

[FLINK-1380] [streaming] Updated SplitDataStream to extend DataStream to get rid of selectAll method for splits

This closes #254.

This closes #293

Before, something like:
def f[T: TypeInformation](data: DataSet[T]) = {
  val tpe = createTypeInformation[(T, Seq[T])]
  println("Type: " + tpe)
}

f(Seq(1.0f, 2.0f)

would fail because the type extractor could not re-use existing
TypeInformation for nested types.

This closes #279

This closes #268

to HadoopFileSystem wrapper, which supports all subclasses of org.apache.hadoop.fs.FileSystem.
This allows us to let users use all file systems with support for HDFS.
The change has been tested with Tachyon, Google Cloud Storage Hadoop Adapter and HDFS.

The change also cleans up the Hadoop dependency exclusions.

Also contains fixes by sewen@apache.org
 - Make MissingTypeInfo optional in TypeExtractor (by default still throws exception)
 - Simplified deferred evaluation of type dependend code by making evaluations lazy
 - Add call location function names to MissingTypeInfo error messages.
 - Improvements on other error messages.

This closes #270

Organized imports for streaming scala examples
Added template parameter for scala streaming iterate
Minor fixes in streaming examples

 - The map driver now also supports this

 - In the merge iterator (merges sorted runs from external sort), we now always use the non-reusing code path,
   because the reusing codepath here implies in all cases additional instances to be held concurrently, and copy
   between elements, which voids the benefits of reusing elements.

 - For many utility iterators (in test cases), consolidates the logic between the two variants of the "next()"
   functions (one calls the other, where possible)

 - Eliminates a few copies between elements in the non-reusing parts (where possible)

 - Removes unused variables in the non-reusing variants (mainly serializers previously used to create instance
   or copy between instances)

 - Remove some unused types

 -  Improves generic type safety (fewer raw types)

This closes #259

This is in preparation for configurable object-reuse mode. We previously
referred to this as mutable object vs. mutable object safe mode or some
such thing.

This closes #285

This closes #286

[FLINK-1363] [runtime tests] Fix race condition in ExecutionVertexCancelTest.testSendCancelAndReceiveFail

This closes #288

[FLINK-1259] [apis] Added comments to interface and docs that filters must not mutate date elements.