Add Staccato tokenizer and retokenize method

[alanakbik]

This PR adds a new tokenizer following the ideas outlined in #3631.

The tokenizer works like any other, but is intended to give a reasonable, if somewhat trigger-happy tokenization of any language. Some illustration of how this works for different languages:

# English sentence tokenized with Staccato tokenizer
print("Testing English with numbers and punctuation:")
sentence = Sentence("It's 03-16-2025", use_tokenizer=StaccatoTokenizer())
desired_tokenization = ["It", "'", "s", "03", "-", "16", "-", "2025"]

for token, desired_token in zip(sentence, desired_tokenization):
    print(f"Got: '{token.text}', Expected: '{desired_token}'")
    assert token.text == desired_token
print("✓ English test passed")


# Test for Cyrillic (Russian)
print("\nTesting Cyrillic (Russian):")
russian_sentence = Sentence("Привет, мир! Это тест 123.", use_tokenizer=StaccatoTokenizer())
russian_desired = ["Привет", ",", "мир", "!", "Это", "тест", "123", "."]

for token, desired_token in zip(russian_sentence, russian_desired):
    print(f"Got: '{token.text}', Expected: '{desired_token}'")
    assert token.text == desired_token
print("✓ Russian test passed")


# Test for Chinese with Kanji/Hanzi
print("\nTesting Chinese with Kanji/Hanzi:")
chinese_sentence = Sentence("你好，世界！123", use_tokenizer=StaccatoTokenizer())
chinese_desired = ["你", "好", "，", "世", "界", "！", "123"]

for token, desired_token in zip(chinese_sentence, chinese_desired):
    print(f"Got: '{token.text}', Expected: '{desired_token}'")
    assert token.text == desired_token
print("✓ Chinese test passed")


# Test for Japanese (mix of Kanji, Hiragana, Katakana)
print("\nTesting Japanese (mixed scripts):")
japanese_sentence = Sentence("こんにちは世界！テスト123", use_tokenizer=StaccatoTokenizer())
japanese_desired = ["こんにちは", "世", "界", "！", "テスト", "123"]

for token, desired_token in zip(japanese_sentence, japanese_desired):
    print(f"Got: '{token.text}', Expected: '{desired_token}'")
    assert token.text == desired_token
print("✓ Japanese test passed")


# Test for Arabic
print("\nTesting Arabic:")
arabic_sentence = Sentence("مرحبا بالعالم! 123", use_tokenizer=StaccatoTokenizer())
arabic_desired = ["مرحبا", "بالعالم", "!", "123"]

for token, desired_token in zip(arabic_sentence, arabic_desired):
    print(f"Got: '{token.text}', Expected: '{desired_token}'")
    assert token.text == desired_token
print("✓ Arabic test passed")

To test this tokenizer, this PR also adds a retokenize method for the Sentence object. Use it like this:

# Create a sentence with default tokenization
from flair.data import Sentence
from flair.tokenization import StaccatoTokenizer

sentence = Sentence("01-03-2025 New York")

# Add span labels
sentence.get_span(1, 3).add_label('ner', "LOC")
sentence.get_span(0, 1).add_label('ner', "DATE")

# Retokenize with a different tokenizer while preserving labels
sentence.retokenize(StaccatoTokenizer())

Further, it adds a typename property for convenience to Label, to allow us to directly access the label type name.