- Added pyproject.toml, following PEP 517/518.
- Slight performance improvement: now use the stack memory for fixed-size temporary buffers in *hiragana_to_katakana()* and *katakana_to_hiragana()*.
- More consistent and precise error handling: assure that *hiragana_to_katakana()* raises a ValueError when *hankaku* option is set to True and the *ignore* optional string contains characters that are not convertible to hankaku katakana. Previously, even though neither 'ゐ' (U+3090) nor 'ゟ' (U+309F) has a corresponding hankaku katakana, only the latter triggered an error.