optimize tsolver

[rmarquis]

Merge SortDepsAur() with FindDepsAur(). These functions are virtually the same but were developed separately to minimize regression and ease testing.

The only purpose of the former is to prepare the tsort list and it can be done at the same time the tree is generated in the latter, resulting in a much faster solver.

[rmarquis]

Note: This isn't straightforward to do, and the deps solving algo might be simplified through refactoring. Have a look at bauerbill, since its solver implementation is very fast and is also the basic of aurutils.

[rmarquis]

Actually, the optimum solution here would be to have topological sorting with grouping, since the dependency tree is solved by "level" rather than by individual element.

I don't think this is possible to achieve with tsort though. Maybe there is a way to emulate that grouping feature through an abstraction layer?

[rmarquis]

Dropping due to #708.