Remove the leftover format placeholder from intword units

[VenishPaneliya]

intword inserts the unit without any % substitution:

singular, plural = human_powers[ordinal]
unit = _ngettext(singular, plural, math.ceil(rounded_value))
...
return f"{negative_prefix}{number} {unit}"

so a printf placeholder left in a translated unit is printed literally. The Bengali, Korean and Vietnamese catalogues still carry %d or %(value)s on the thousand…googol units:

humanize.i18n.activate("bn_BD"); humanize.intword(1_500_000_000)
humanize.i18n.activate("vi_VN"); humanize.intword(1_200_000)
humanize.i18n.activate("ko_KR"); humanize.intword(1_200_000)

On main:

'1.5 %d বিলিয়ন'
'1.2 %(value)s triệu'
'1.2 %(value)s million'

With this PR:

'1.5 বিলিয়ন'
'1.2 triệu'
'1.2 million'

This strips the placeholder and the separator that follows it (a plain space or U+00A0) and leaves the translated words untouched.

Scope is deliberately narrow: only the twelve intword power msgids are changed. Placeholders elsewhere in these catalogues are substituted normally and are left alone — naturaldelta's "%d day" and friends still need theirs. After the change, none of the 36 catalogues emits a placeholder for any of the twelve powers.

Added a parametrised regression test over the three locales, checking that intword never returns a % at any power. It fails 3/3 on main and passes here.

Two notes I did not act on, since both need a native speaker rather than a scripted edit:

• ko_KR renders several powers as the English words (million, billion), and its trillion entry reads billion.
• ko_KR and vi_VN also contain a couple of French leftovers (décillion, gogol).

Happy to split those out separately if you would like them addressed.