Better support for BatchEncoding

[jss367]

. Example:
texts = ["this is my test text", "this is another test text"] * 100

# load a pre-trained tokenizer
tokenizer = BertTokenizerFast.from_pretrained(
    "bert-base-uncased",
    add_special_tokens=True,
    max_length=input_dimensions,
    pad_to_max_length=True,
    return_tensors="pt",
)

encoded_text = tokenizer(texts, padding=True, truncation=True, return_tensors="pt")