Add Ukrainian, Vietnamese and Turkish translations

[Nachtalb]

Adds three more UI languages: Ukrainian (uk), Vietnamese (vi) and
Turkish (tr). Catalog count goes 15 → 18.

Unlike the previous batch, these were picked from actual usage rather than a
world-population ranking. A read-only query against the production
users.language_code column (6,439 users) put them as the three largest locales
with no catalog:

code	users	share
uk	56	0.88%
vi	40	0.63%
tr	34	0.53%

Everything below these drops to ko 15, ms 13, nl 13, th 11 and then single
digits across 40 distinct codes, so this is where the returns flatten out.
Coverage of users with a known language goes from ~96.3% to ~98.3%.

Purely additive — no behaviour change. Each language is one TOML catalog plus
one entry in _LANG_NAMES for the /settings picker; auto-detection off
Telegram's language_code starts working immediately.

Verification

Each catalog was checked against en.toml for identical key sets and ordering,
identical {placeholder} sets per key, identical HTML tag sequences, matching
\n counts and no untranslated leftovers. _LANG_NAMES and the shipped
*.toml files are asserted to be the same set.

Full local gate green: ruff check, ruff format --check, ty check,
pytest (558 passed).