Split sourced-ml package to algorithms and data collection parts

[zurk]

Dependent projects such as https://github.com/src-d/style-analyzer need only algorithms part of the sourced-ml: https://github.com/src-d/ml/tree/master/sourced/ml/algorithms

Data collection part uses deprecated jgit-spark-connector which depends on old packages. This leads to unpleasant dependency conflicts: https://github.com/src-d/style-analyzer/pull/719/files#diff-354f30a63fb0907d4ad57269548329e3R30

That is why we should split the package into two parts.

[Guillemdb]

I am currently trying to make sense of the src-d/ml package, and I am building a map of how the different files depend on each other. I will be updating it during this week.

I hope that when it's finished it helps splitting the package in two.

src-d/ml files

Edit link

The different colors mean the following:

• Blue: name of the module
• Green: Files that do not depend on bblfsh
• Yellow: Files that depend on bblfsh, but only only to use features of a UAST that "should be stable". (Role type, parents of a node, children, etc.)
• Orange: Files that depend on bblfish that may need changes when using graphs instead of trees.