Cluster label adaptation speed-up

[francois-drielsma]

Updates geared towards faster training times:

• Completely revamped the cluster label adaptation routine into its own class (x4 speed-up)
  • Rewrote local numba implementation of union_find based on a faster radius_graph computation
  • Rebased local numba implementation dbscan on the above two for significant time-saves
  • Parallelized cluster breaking with the above
• Minor bug fix in Particle initialization
• Minor bug fix in larcv_check_valid.py script