Add IndexMap imbalance "diagnostic" function

[IgorBaratta]

This PR adds a new imbalance() function to the IndexMap class.
The imbalance is defined as the maximum number of indices on any process divided by the average number of indices per process, and is computed separately for owned and ghost indices.

Usage example:

import ufl
import dolfinx
from dolfinx import fem
import numpy as np
from dolfinx.io import XDMFFile
from mpi4py import MPI

comm = MPI.COMM_WORLD

# Create a mesh and function space.
mesh = dolfinx.mesh.create_unit_cube(comm, 10, 10, 10,
                                     cell_type=dolfinx.mesh.CellType.tetrahedron,
                                     ghost_mode=dolfinx.mesh.GhostMode.none)
V = fem.FunctionSpace(mesh, ("Lagrange", 5))
index_map = V.dofmap.index_map
tdim = mesh.topology.dim

dof_imbalance = index_map.imbalance
cell_imbalance = mesh.topology.index_map(tdim).imbalance

if (comm.rank == 0):
    print("dof imbalance", dof_imbalance)
    print("cell imbalance", cell_imbalance)

dof imbalance [1.0140896035461473, 1.3187465232709068]
cell imbalance [1.006, -1.0]

[chrisrichardson]

So, perfect balance = 1? And it can never be less than 1. I am not sure about the -1, maybe this should also just be 1 - because zero everywhere is also perfectly balanced?

And then, why not just subtract the 1? In the example above, we are mostly interested that one process has 31.8% more ghost dofs than average... 0.318 would be the "imbalance", 0.0 perfect balance...

[chrisrichardson]

It would be useful to say that "1.0" represents a "perfect" balance.

[jhale]

This seems useful, and is non-trivial to implement. Thanks!