Support gremlin.spark.skipPartitioning configuration.

If a VertexProgram does not use message passing, then its best to NOT partition after load as its pointless to do so.

In particular, for TraversalVertexProgram, if the submitted traversal does not contain a VertexStep then partitioning can be avoided. This can be reasoned via a SparkPartitionStrategy, but for now, simply making the configuration and having it do its job is sufficient.