Improve biharmonic demo

[jorgensd]

Rework the biharmonic demo added by @juliusgh to include:

• Skip single precision runs. The matrix is very poorly conditioned and does not converge to the same solution for single and double precision (this also holds for the existing demo).
• Better choice of element (P2 yields sub-optimal convergence)
• Usage of clamped boundary conditions, rather than simply supported bcs
• Explain the difference between the bcs.
• Use a method of manufactured solution as a verification mechanism
• Add relevant references to papers on discretization.