Bug fix - alias remapping.

Pre-made tokenizers, CNN- and transformer-style encoders + training, …

…BLOSUM-based augmentation, and flatten_swiss, which extracts tabular data for NLP pretraining

Training for autoregressive language modeling of biological sequences.

Numpy type-parity, trainable sparse softmax, and a SeqEncoder factory…

… for biological sequence embeddings.

Add is_padded(), includes_bos(), includes_eos(), alphabet_size(), pad…

…(), eos(), bos() functions to Tokenizers; add prebuilt Tokenizers, and a bioseq.make_embedding utility to handled padding and torch.nn.Embedding creation.

v0.1.2 - add direct tokenizers which can be used in conjunction with …

…torch.nn.Embedding.

Expanded alphabet usage