Added RobertaTokenizer

[jkrukowski]

This PR adds RobertaTokenizer.

It was tested against the python transformers implementation with the following code:

from transformers import AutoTokenizer

model_path = "FacebookAI/roberta-base"
# tested as well with ibm-granite/granite-embedding-30m-english

tokenizer = AutoTokenizer.from_pretrained(model_path)
assert tokenizer.tokenize("l'eure") == ["l", "'", "e", "ure"]
result = tokenizer(["l'eure"])
assert result['input_ids'][0] == [0, 462, 108, 242, 2407, 2]

assert tokenizer.tokenize("mąka") == ["m", "Ä", "ħ", "ka"]
result = tokenizer(["mąka"])
assert result['input_ids'][0] == [0, 119, 649, 5782, 2348, 2]

assert tokenizer.tokenize("département") == ["d", "é", "part", "ement"]
result = tokenizer(["département"])
assert result['input_ids'][0] == [0, 417, 1140, 7755, 6285, 2]

assert tokenizer.tokenize("Who are you?") == ["Who", "Ġare", "Ġyou", "?"]
result = tokenizer(["Who are you?"])
assert result['input_ids'][0] == [0, 12375, 32, 47, 116, 2]

assert tokenizer.tokenize(" Who are you? ") == ["ĠWho", "Ġare", "Ġyou", "?", "Ġ"]
result = tokenizer([" Who are you? "])
assert result['input_ids'][0] == [0, 3394, 32, 47, 116, 1437, 2]

assert tokenizer.tokenize("<s>Who are you?</s>") == ["<s>", "Who", "Ġare", "Ġyou", "?", "</s>"]
result = tokenizer(["<s>Who are you?</s>"])
assert result['input_ids'][0] == [0, 0, 12375, 32, 47, 116, 2, 2]

[FL33TW00D]

Sorry for the delay reviewing this, looks brilliant great work!