Issues while adding support for extruded meshes

An incomplete list of issues I encountered while implementing extruded meshes:

Issues to be considered during structured interface adoption:

• iterate over dune kernels and get the correspondend integrand instead of the other way around
• clean up accumulation and coefficient access
• make coefficient access part of the visitor
• compute normals for multilinear column
• decide what quadrature_position should return (position on extruded face or base face)
• use known normals for horizontal faces
• iterate over dune kernels and get the correspondend integrand instead of the other way around
• clean up accumulation and coefficient access
• make coefficient access part of the visitor
• two ways of handling horizontal faces:
  • generate one kernel with predicates for face instructions (maybe during visiting)
  • generate separate kernels for each integral type and merge them <-- current approach

Open issues:

• using first element in mesh for filling quadrature cache -> use element type instead
• use explicit geometry type instead of using geometry from given entity
• no support for UFL tensorelements (probably not needed)
• define basis type explicitly in lop
• use explicit basis in cache evaluation
• literal c code is hard to adjust -> mixing "c++" objects with symbolic objects is tedious
  • I hope this will get fixed with #159

I don't think these are controversial topics, so I would also fix them in a more unified way when I work on the structured interface adoption.