analyzer scheduling could be more effective

[pietdevaere]

Lets say that analyzer A has to run 12 times, and that analyzer B will process the output of these 12 runs in 1 run.
Currently, it sometimes happens that analyzer B is executed before analyzer A has ran 12 times.
This causes analyzer B to run multiple times to come to the same results as when analyzer A would have ran 12 times first, and then analyzer B was ran.

This is not very efficient.

A possible solution would be to construct some sort of tree of which analyzers use which analyzers output, and then schedule the analyzers by traversing this tree BFS.