add concurrent support for removeDeletedNodes

[jbellis]

This is important to allow larger indexes to support deletes without pausing updates and searches.

Unfortunately my first attempt at solving this didn't work out and was removed in #273.

The problem is that the logical clock used to detect when searches were done does not work when GraphSearchers are pooled and reused (which we want to allow to reduce allocation churn).

I think the solution is as simple as just creating a new View per search (the Views are much more lightweight than the Searchers), but otoh it's a bit of a shame to churn all Views when it's only the on-heap one that is a problem.

[jbellis]

closing in favor of #272