Fix Module.save : serialise loss & writer

[opeltre]

TODO:

• Write to a dictionnary of lists with Module.write by default
• Test when overwriting with tensorboard methods, e.g.

  module.write = summary_writer.add_scalar

[opeltre]

fix #16

[opeltre]

Dans unshift_flows.py essayer:

writer = SummaryWriter(args.output)

def write (name, val, n_it): 
    writer.add_scalar(name, val, global_step=n_it)

module.write = write

A essayer mais moins sûr que ça marche (si on appelle bien module.write avec les bons arguments dans module.fit):

writer = SummaryWriter(args.output)
module.write = writer.add_scalar

[opeltre]

Solves #16