Fix #158

[wravery]

There's a race condition in Schema::WrapType where it caches the WrapperType in a map, keyed off of the original type. If you perform query validation on multiple threads, they may simultaneously lookup and emplace new values in the cache, invalidating the iterators on one of them, possibly resulting in duplicate entries.

I'm using a std::shared_mutex as a reader/writer lock on each of these maps. It can find a match with a std::shared_lock, and if there is no match, it'll trade that for a std::unique_lock and perform the emplace, which is equivalent to a find if the key was added on another thread before acquiring the std::unique_lock.