toRomaji doesn't respect customRomajiMapping if the input is katakana

[fasiha]

This works: toRomaji('つじぎり', { customRomajiMapping: { じ: 'zi', つ: 'tu', り: 'li' } }), and produces tuzigili per the documentation.

However, this doesn't work? toRomaji('ツジギリ', { customRomajiMapping: { ジ: 'zi', ツ: 'tu', リ: 'li' } }) produces tsujigiri.

(I don't want to pass all input to toRomaji first through toHiragana and use hiragana-only customRomajiMapping because that changes how long vowels are handled.)