Make the trainer optional when training

[n1t0]

Current state

When training a Tokenizer, we must provide a Trainer:

trainer = BpeTrainer(...)
tokenizer.train(files=[ ... ], trainer=trainer)

This is true even if we instantiate the BpeTrainer without changing any defaults.

Goal

Makes the Trainer optional:

tokenizer.train(files=[ ... ])

How

The Model should be able to return a default trainer:

trainer = model.get_trainer()