Python improvements

[n1t0]

• Uniformize lowercase vs do_lowercase in the various implementations
• Formatting is now done through Black.
• CI now checks that the python bindings build and that the code is well-formatted, both in Python and Rust.
• Fix Usage examples (pypi and readme) are outdated [BPETokenizer rename] #152 with 5daf1ee
• Fix [bug] The parameter add_special_tokens has no effect when set to False in Bert WordPiece tokenizer #141 with d48fdbe

@mfuntowicz You may have some things to change in huggingface/transformers#2674 to respect 1509f74

[julien-c]

👍 for black formatting!