Support for fastText embeddings

• Add support for reading fastText embeddings through the ReadFastText trait.
• Generalization of SubwordVocab for different indexing schemes.
• Add an iterator over both n-grams and their subword indices (NGramsIndicesIter).
• Support similarity queries for embeddings (EmbeddingSimilarity). The word similarity trait was renamed from Similarity to WordSimilarity.
• Better and more informative errors on I/O errors.
• Use library-specific error types.
• The words for analogy queries are now accepted as an array of length 3.
• Make the chunk modules (norms, metadata, vocab, and storage submodules of the chunks module).