Shared Voronoi functionality between baseplate and powder init

[MattRolchigo]

An edge case was previously possible in ExaCA where, due to the fact that the powder layer initialization left some cells with a "zero" GrainID, simulations of FromFile and FromFinch problems could get locked in a loop where no valid grain could claim the liquid cells in some melt pools for many time steps, often leading to simulation timeout. It also lead to unnatural grain shapes as grains stretch across multiple melt pools.

This changeset uses Voronoi helper functions to initialize both the baseplate (which previously used a Voronoi tessellation to assigned GrainIDs to cells) and each powder layer, so that no cells have a GrainID of 0 following initialization. The PowderDensity input is replaced with a MeanPowderGrainSize input, and for consistency, the MeanSize input is renamed MeanBaseplateGrainSize. Minor changes to the std::cout statements are also made to log which GrainID values are being used for which parts of the simulation (baseplate, powder layers, nucleated grains)