Cache Knowledgegraph

[yuancu]

To accelerate inference, we can cache a subset of edges and entities.

• Create CachedKnowledgeGraph that loads cached triplets
• Create a script that caches triplets according to a set of entities.