Optimize schema retrieval in HortonworksSchemaRegistry

If HortonworksSchemaRegistry is used by a processor with multiple threads (Concurrent Tasks > 1) and no schema version has been specified (so the 'latest' one needs to be retrieved), then the threads refresh the NiFi side schema cache in parallel / multiple times. Subsequent threads should block until the first one retrieves the schema from SchemaRegistryClient.

Proposed solution: use a cache implementation that correctly handles parallel get() requests (eg. caffeine).