Lattice issues

[mjpost]

Joshua truncates sentences to 200 tokens, but turns them into a lattice first and computes the shortest path between all pairs. This causes huge problems for buggy input (e.g., https://groups.google.com/forum/#!topic/joshua_support/-2PykKqbnU4 ). A couple things that should change:

• Don't apply an O(n^2) algorithm to buggy input. Maybe only call calculateAllPairsShortestPaths() lazily.
• Compute it more efficiently (for example, if you know the input is just a sentence, you could add some shortcuts)