mhcflurry 1.1

[timodonnell]

• new more accurate models that are ensembles of 45 different architectures
• support for training losses that make use of inequalities
• incorporation of mass-spec training data and downloadable models that are trained both with and without mass spec data
• parallelized training and percentile rank calibration
• faster model loading: lazy load model weights
• progress bars