Not all variants appear in YaspGrids. The number of variant
reduces from 4*d*d -> 2*d for skeleton integrals.

Only 6 orders of magnitude to go!

Easier access expressions. No overhead (I hope so)

Vectorization as preprocessing

Refactoring of the entire visiting process:
* Have a dry run that collects information
* Directly construct large matrices during tree visiting
* Allow code generation for terminals to also handle indexing into the terminal!

See merge request !71

Feature/dune opcounter

See merge request !70

It caused quite a bit of trouble with reuse of large matrices.

Bump VCL

See merge request !73

by storing them in the visitor

Bump loopy

See merge request !72