9x speedup of LArCellMonAlg
Move LArCellMonAlg to use Monitored::Collection for most calls instead of Monitored::Scalar; it is no longer the slowest DQ algorithm running on ESD. Also restructured the monitoring group structure for a bit more clarity. Output is identical to previous version.
Attn @pavol (in particular is there a smarter way to determine how many cells there are in each layer so we don't need to reserve too much memory?)