Fix Japanese tokenizer URL handling

[bhushanasati25]

Fixes#13928.

Description

The Sudachi-backed Japanese tokenizer now protects URL spans before tokenizing the surrounding text, so URLs such as https://example.com/jp stay as one token in Japanese sentences. The tokenizer uses spaCy's shared URL matcher for validation, preserves doc.text_with_ws, and marks the resulting token with the normal lexical LIKE_URL attribute.

Types of change

Bug fix.

Tests

• python -m pytest -q spacy/tests/lang/ja/test_tokenizer.py -k keeps_urls
• python -m pytest -q spacy/tests/lang/ja/test_tokenizer.py -k 'not sub_tokens'
• python -m ruff format spacy/lang/ja/__init__.py spacy/tests/lang/ja/test_tokenizer.py --check
• python -m ruff check spacy/lang/ja/__init__.py spacy/tests/lang/ja/test_tokenizer.py --select I
• git diff --check

Note: python -m pytest -q spacy/tests/lang/ja/test_tokenizer.py fails one existing test_ja_tokenizer_sub_tokens expectation with the current sudachidict_core==20260723, where the first two subtoken entries are [[ ]] instead of None. The new URL regression passes.

Checklist

• I confirm that I have the right to submit this contribution under the project MIT license.
• I ran the tests, and all new and existing tests passed except for the unrelated SudachiDict-version-sensitive case noted above.
• My changes don't require a change to the documentation, or if they do, I've added all required information.