quantity.get_vertex_values is slow

[anonymous]

Profiling of okushiri (yieldstep=0.05, finaltime=2) gave the following runtimes:

*store_vertex_values_uniquely(True): 138.3s *store_vertex_values_uniquely(False): 505.6s *No storing: 28.2s

Profiling revealed that quantity.get_vertex_values is the culprit.

I suggest rewriting some of it in C and perhaps always use mean as the reduction operator. We don't really need any other.

This problem is worst in simulations with frequents storing.