Decompose computations for product Dynkin types

[pbelmans]

Whilst finishing the release for v1, I realized that decomposing computations for product Dynkin types could be a big win. Not in how readable the code is, but performance-wise. This is confirmed by b544188
.

This is a rather hefty change to still be included in v1, so let's keep this for v1.1, to be released soon.

• add more real-life examples from algebraic geometry to the tests and benchmarks
• look for ways to not regress on small product types (maybe only do this if total Dynkin rank is above a certain threshold?)

[pbelmans]

Some quick experiments in PartialFlagVarieties.jl don't give a major speedup, despite a significant speedup being visible in the artificial examples.