* Reimplemented connection tests to work with `RKConnectionDescription`. Introduced new `RKConnectionTestExpectation` class
* Slimmed down API by removing proxy methods on `RKMappingTest`
* Eliminated `verifiesOnExpect` behavior in favor of using `evaluateExpectation:`
* Ensured compatibility with RKKiwiMatchers
* Expands test coverage for the `appropriateObjectRequestOperationWithObject:method:path:parameters:`
* Uses an object graph visitor to completely navigate the mapping graph, ensuring that an `RKEntityMapping` appearing at any nesting level will be correctly handled
* Migrated the caches to return `NSSet` to eliminate duplicate objects when your cache keys overlap
* Introduced new recursive strategy for building cache keys in `RKInMemoryManagedObjectCache`
* Added support for array cache key values in `RKFetchRequestManagedObjectCache`
* Re-enabled a slew of tests that were disabled during 0.20 development
* Replaces primary key with `RKEntityIdentifier`
* Add support for use of compound keys for object identification
* Refactor `RKConnectionMapping` to `RKConnectionDescription` and add support for connecting with multiple attributes
* Clarify naming of representation key methods to better match naming conventions
* Add type transformation support for object identification
* Greatly expand test coverage for object identification
* Drop the `NSEntityDescription` category
* Simplify the `RKManagedObjectCaching` protocol
* Add compound key support to the Fetch Request and In Memory Cache implementations
* Replace Kiwi with Specta for tests where contexts are helpful for organization
* Rename `defaultValueForMissingAttribute` to `defaultValueForAttribute`
* Renamed RKObjectPaginator to RKPaginator for simplicity
* Removed delegate on paginator in favor of simple success/failure blocks
* Added `willMapDeserializedResponseBlock` for manipulating the response body before mapping begins
* Added support for cancelling an in-progress pagination request
* Supports replacing the RKSearchWord fetch strategy to enable caching for performance
* Supports declining creation of search words
* Supports declining of indexing for specific objects
* Supports notification when objects are indexed and search words are added to the index