Add a layer of caching to NLP server methods

[gcampax]

To reduce latency and compute costs, we should briefly cache all NLU and TTS requests in some in memory caching system, like a Redis or memcached.