Improve field energy integration for electrostatic and magnetostatic simulations

[sebastiangrimberg]

Use the scalar voltage (H1) and vector potential (Nedelec) fields, which typically have smaller space dimension than the corresponding electric field (Nedelec) and magnetic flux density (Raviart-Thomas).

This can actually become a bottleneck for large capacitance or inductance matrix sizes, where O(N^2), with N the number of terminals, energy calculations are needed.

[sebastiangrimberg]

A brief performance assessment:

• Mesh: 56.7M elements, mixed tet/prism
• p = 4: 658M H1 dofs, 2.46B ND dofs
• N = 258 terminals (cap. matrix is N x N, require N * (N+1) / 2 energy calculations)

main (min/max/avg time in seconds on 4096 cores):

Postprocessing                3099.500    3099.783    3099.548

This PR:

Postprocessing                1060.840    1061.117    1060.878