Investigate potential benefits of replacing hash-to-index mappings with Rcs

We want to investigate the potential to replace the newly-implemented mappings of hashes to usize indices with Rc pointers to the hashes.

It's hoped that this might have a smaller impact on the code, while not costing us a significant drop in performance.

We need to consider whether to try to replace one of the exiting instances of such a mapping (probably the one relating to Event s would be simpler to use than the PeerId one) in order to examine the results, or whether to try to create a separate throwaway crate which will need to roughly replicate the behaviour of Parsec wrt such a mapping (i.e. have similar numbers of elements and similar ratio of insertions/accesses/deletions)