Patch from Thomas Graves to improve the YARN Client, and move to more production ready hadoop yarn branch

Fix review comments, add a new api to SparkHadoopUtil to create appropriate Configuration. Modify an example to show how to use SplitInfo

cache sizes due to having 2 instances of the Scala compiler and a bunch
of classloaders.

Move back to 2.0.2-alpha, since 2.0.3-alpha is not available in cloudera yet. Also, add netty dependency explicitly to prevent resolving to older 2.3x version. Additionally, comment out retrievePattern to ensure correct netty is picked up

Added TwitterInputDStream from example to StreamingContext. Renamed example TwitterBasic to TwitterPopularTags.

As part of this, changed our Scala 2.9.2 Kafka library to be available
as a local Maven repository, following the example in
(http://blog.dub.podval.org/2010/01/maven-in-project-repository.html)