System tests should check convergence rates

At the moment we have many system tests, but many of them are static and we have run into issues where the L2 error may be below a threshold in one case, but when testing convergence rates may not be at optimal of ele degree +1.

Start by inspecting the existing system tests and how they compute L2 errors and finite-element degree. Identify representative tests where convergence rates can be measured, then define done as checks showing the expected degree + 1 convergence rate rather than only a single error threshold.

Written by the indexing model from the issue text.