pykakasi fails to romanize half-width katakana with voiced marks

pykakasi does not romanize half-width katakana correctly, particularly when a half-width voiced or semi-voiced mark (U+FF9E / U+FF9F) follows the base kana.

Convert half-width katakana such as a half-width ka followed by a half-width dakuten; the output is wrong instead of 'ga'.

NFKC-normalize half-width katakana and combining voiced marks to their full-width equivalents before romanization.