performance : allow to have labels in cache

Computing labels positions and ordering takes time for large graphs even when the number of available labels is low.

The user experience would be much improved if we had the possibility to put the result of labels computation in cache for reuse across same different displays of the same graph.

Start by locating the code that computes label positions and ordering, then inspect how separate displays of the same graph are created and whether existing caching patterns apply. Done means the label computation result can be reused across displays of the same graph and large graphs no longer repeat that work unnecessarily.

Written by the indexing model from the issue text.