Implement Hadoop DSL Canonical Compiler and Azkaban Reverse Job Compiler

[convexquad]

The idea is to implement a Hadoop DSL compiler that actually writes out formatted Hadoop DSL. This would be necessary for a Azkaban job files -> Hadoop DSL reverse compiler.

In people's Hadoop DSL files they will have all kinds of Groovy that is actually evaluated at Gradle build time. They end up with a Hadoop DSL structure in memory. The canonical compiler would spit out the Hadoop DSL structure as Hadoop DSL itself, but obviously with no Groovy logic in it.

Once the canonical compiler is in place, the task of implementing an Azkaban reverse compiler would simply consist of a processor for the Azkaban .job and .properties files that builds up the corresponding Hadoop DSL structure in memory and then just writes it out using the canonical compiler.

[convexquad]

The reverse compiler would be a more challenging longer-term task. It should definitely be broken up into these two parts.