Set up tests

[thatbudakguy]

• Pull out MockCupyNdarray for testing
• Remove shared state in tests
• Add some more initializer tests
• Add a default model config for Phonemizer
• Use predicted data when initializing model
• Update empty check in examples_to_truth
• Change numpy import
• More explicit get_aligned_phonemes implementation
• Port Doc.to_array for phonemes and test it
• Add attribute registry convenience function
• Pull MockCupyNdarray up to top level
• Make to_phonemes_array return a numpy array
• Adapt Example.get_aligned for phonemes
• Add more extension attr getters; consolidate tokens module
• Flatten training module
• Use convenience methods in phonemizer tests
• Refactor phonemizer to match tagger more closely
• Remove unused test dir
• Flatten pipeline directory
• Extract scorer to new module
• Ignore OSX thumbnail info
• Add info about test coverage
• Prevent phoneme values from being None