Try serializing job details instead of using a file on HDFS

[jphalip]

Currently we pass some information between the job's tasks and processes using the JobDetails class and its corresponding serialized version as a file stored on HDFS.

Instead, we could perhaps store those details in the HiveBigQueryConfig class, which is serializable. For reads it could be passed via the BigQueryInputSplit.

For writes, I'm not sure yet as BigQueryOutputFormat doesn't seem to have a consistent way or passing custom serialized information besides the JobConf... To be investigated.

As part of this, we should revisit the way we're using the Guice injector. Ideally that would only be used once at the beginning, instead of using it multiple times in multiple contexts as we do currently.