Refactor

[LLParse]

• Rework configuration system to use typesafe snakeyaml api
• Minimize dependency scope, remove unused dependencies for ~50% size savings in built artifacts
• Define protocol buffer messages for journal, name, data nodes
• Remove configuration logic from artifact server, use Mesos data fields to pass protobuf between scheduler & executor (this fixes idempotency [WIP, name/data nodes not quite finished])
• Refactor/simplify executor system, emit hdfs-site.xml using {{mustache}} template and task protobuf data
• Pool enough offers to launch journalnodes all at once (TODO: save journal rpc addresses in persistent state, use ports in offer when configured ports are set to '0', genericize pooled offer logic to work for name/data node acquisition phases, wrap offer system in timer to prevent holding onto resources indefinitely in too-few-resources scenarios)
• TODO: major scheduler refactor, tweak name/datanode config

[elingg]

Instead of using a yaml config here, what I would prefer to do is something similar to the following PR (which didn't get finished), mesosphere-backup#123. I would like to use environment variables or command line arguments (similar to Cassandra). These values could then be used to generate the *-site.xml configs used by HDFS.

[elingg]

Overall comment, this would need to be rebased with mesosphere/master