Improve gradle project meta inference

[Strum355]

Currently importing as an EclipseProject model. Importing as an IdeaProject (preliminarily) appears to be better at providing dependencies locations (which are fed into the -classpath flag). More investigation needed, maybe results will be merged?

[Strum355]

IdeaProject model is less reliable for getting java source versions (requires the use of JavaVersion.XYZ), while EclipseProject parses string if provided e.g. "1.8"

[Strum355]

Combining the discovered classpath/dependencies list appears to have a net positive effect on the number of javac errors. Still generally a large number of symbols and packages not resolved (looking at you, org.apache.thrift)

[Strum355]

Exploring Kythe-style javac interception in #78
Has the unfortunate effect of requiring the entire build phase to be run (without gradle caching), but may be a necessary effect to get classpaths completely correct