Make building PLDB via ScrollHub 400% faster

<>

If we keep the Scroll Files in memory, we can build in 80% less time. In my tests on my macbook went from 33s to 6s.

Steps:

  1. cleanup all the fusion code, eliminating concepts, simplifying, testing.
  1. add memory persistence in scroll/scrollhub, properly handling writes.
  1. build from memory. first build will still be slow, but then we'd get 4x faster builds.

seems to be the biggest possible win right now.