Speed up Tokenization by optimizing cast_to_python_objects

[lhoestq]

I changed how cast_to_python_objects works to make it faster.
It is used to cast numpy/pytorch/tensorflow/pandas objects to python lists, and it works recursively.

To avoid iterating over possibly long lists, it first checks if the first element that is not None has to be casted.
If the first element needs to be casted, then all the elements of the list will be casted, otherwise they'll stay the same.
This trick allows to cast objects that contain tokenizers outputs without iterating over every single token for example.

Speed improvement:

import transformers
import nlp

tok = transformers.BertTokenizerFast.from_pretrained("bert-base-uncased")
txt = ["a " * 512] * 1000
dataset = nlp.Dataset.from_dict({"txt": txt})

# Tokenization using .map is now faster. Previously it was taking 3.5s
%time _ = dataset.map(lambda x: tok(x["txt"]), batched=True, load_from_cache_file=False)
# 450ms

# for comparison
%time _ = tok(txt)
# 280ms

[thomwolf]

Nice! Maybe we can add a few tests on these behaviors?

[lhoestq]

I took your comments into account and added tests for cast_to_python_objects