Sense inventory caches should use Guava's caching API

[logological]

Originally reported on Google Code with ID 55

Some of our sense inventories, such as wordnet, cache senses, since their descriptions
and other data can be expensive to compute.  The caches are manually coded rather naïvely.
 It would probably be better to use something like Guava's caching API, whose caches
can be memory-limited.

Reported by [email protected] on 2013-12-04 12:58:38