For now, this only adds MarshalSerializer, but it lays the groundwork
for other supporting custom serializers.  Many of these mechanisms
can also be used to support deserialization of different data formats
sent by Java, such as data encoded by MsgPack.

This also fixes a bug in SparkContext.union().

If we support custom serializers, the Python
worker will know what type of input to expect,
so we won't need to wrap Tuple2 and Strings into
pickled tuples and strings.

Write the length of the accumulators section up-front rather
than terminating it with a negative length.  I find this
easier to read.

Currently PythonPartitioner determines partition ID by hashing a
byte-array representation of PySpark's key. This PR lets
PythonPartitioner use the actual partition ID, which is required e.g.
for sorting via PySpark.

Passing large volumes of data through Py4J seems
to be slow.  It appears to be faster to write the
data to the local filesystem and read it back from
Python.

Objects serialized with JSON can be compared for equality, but JSON can be slow
to serialize and only supports a limited range of data types.