Fix 500s

• catch subnode read errors
• log errors when they happen
• log more
• force graph refresh if inconsistent?
• try other cache setups
  • TLDR: cachetools.func seems to be much faster than Flask's built in SimpleCache, and even than Redis.
  • This might have something to do with the fact that I'm caching the graph as a monolithic entity, and it's already 50MB of data. That's a lot to copy/transfer over a socket.