Pysparkling Versions Save

• make dependencies optional: boto, requests
• compatibility

testing continuous deployment

• Broadcast, Accumulator and AccumulatorParam by @alexprengere
• support for increasing partition numbers in coalesce and repartition by @tools4origins

• fixes for HDFS thanks to @tools4origins
• fix for empty partitions by @tools4origins
• api fixes by @artem0 and @tools4origins
• various updates for streaming submodule
• various updates to lint and test system
• logging: converted some info messages to debug
• ... documentation for some point releases is missing

• retries for failed partitions
• improve pysparkling.streaming.DStream
• updates to docs

• major addition: pysparkling.streaming
• updates to RDD.sample()
• reorganized scripts and tests
• added RDD.partitionBy()
• minor updates to pysparkling.fileio

small improvements to fileio and better documentation

• reimplement RDD.groupByKey()
• clean up of docstrings

• faster text file reading by using io.TextIOWrapper for decoding

* Google Storage file system (using ``gs://``)
* dependencies: ``requests`` and ``boto`` are not optional anymore
* ``aggregateByKey()`` and ``foldByKey()`` return RDDs
* Python 3: use ``sys.maxsize`` instead of ``sys.maxint``
* flake8 linting