fix(python): preserve async encode character offsets

[gry67673905]

Summary

• Keep Tokenizer.async_encode consistent with Tokenizer.encode for Python string inputs by preserving character-based offsets.
• Add a Unicode regression for "é a", where UTF-8 byte offsets differ from Python character offsets.

Validation

• python -m pytest -q -p no:cacheprovider tests/bindings/test_tokenizer.py -k test_async_encode_char_offsets — 1 passed, 48 deselected
• python -m pytest -q -p no:cacheprovider tests/bindings/test_tokenizer.py::TestTokenizer — 30 passed (two existing Python 3.14 fork() deprecation warnings)
• cargo fmt --manifest-path bindings/python/Cargo.toml -- --check
• ruff check tests/bindings/test_tokenizer.py
• ruff format --check tests/bindings/test_tokenizer.py