Clean up collection_local_send

[lifflander]

We would like to clean up the collection_local_send performance test. Let's create one collection element per rank and then make sure we are sending locally from each rank to the element. We can query the map function or ask VT which elements are local to do this.

          @JacobDomagala @lifflander I'm trying to understand what the `collection_local_send.cc` perf tests actually do. Is it true that only rank `0` runs the timed perf test, and only does a send to collection element `0`, which is assumed to be local to rank `0` but not verified? It seems like some stuff could be simplified, like (a) consistently using `my_node_` instead of `theContext()->getNode()`, (b) also only pre-allocating on rank `0`, and (c) using `preallocate_` to conditionally pre-allocate based on what happened in `initialiaze` instead of having the separate `perfMakeRunnable` and `perfMakeRunnablePreAllocate` calls.

Originally posted by @nlslatt in #2097 (comment)