Optimise implicit radiation solving

[themikelau]

Summary of changes:

• Iterate only over subset of particles that have not converged.
• Lower tol_rad from 10^-5 to 10^-6 in test suite so that calculation of dxi/dt with implicit radiation is accurate enough when iterating over individual particles.
• Move allocation of compacted neighbour list and neighbour arrays into new module implicit (utils_implicit.f90).
• Only allocate arrays if unallocated, as opposed to doing so every time step.
• Remove calculation of redundant dvdx in fill_arrays (it is identical to the calculation in density). Remove check for dvdx in the test suite.
• Put calculation of flux limiter lambda and Eddington factor into the compute_flux particle loop. Remove timing for computing lambda and eddington.
• Cache cv, lambda, eddington, and opacity in EU0(6,npart) array.