Add Estonian country name translations

[timujinne]

Problem

BeamLabCountries.Translations ships 15 locales — ar de en es fr it ja ko nl pl pt ru sv uk zh — and Estonian is not one of them. An Estonian-facing host gets nil from get_name(code, "et") and has to fall back to the English name, so its country dropdown reads "Estonia" on an otherwise Estonian page.

Change

• priv/data/locales/et.json — the same 249 codes the other locale files carry, sourced from CLDR (cldr-localenames-full/main/et/territories.json), the same lineage as the existing locale files.
• "et" registered in @supported_locales (now 16), moduledoc, README example and CLAUDE.md data-location note updated to match.
• A test that Estonian covers every country English covers.

Translations.get_name("EE", "et")  # "Eesti"
Translations.get_name("FI", "et")  # "Soome"
Translations.get_name("GB", "et")  # "Ühendkuningriik"

Note on the coverage test

It compares Estonian against English, not against BeamLabCountries.all/0. The locale files skip a few historical codes that the country data still carries (AN, Netherlands Antilles), so an all-countries assertion fails for every locale, not just this one. Estonian is held to the same bar as the rest.

Verification

$ mix test
121 doctests, 160 tests, 0 failures

$ mix format --check-formatted
(clean)

CHANGELOG.md and @version are left to the maintainer.

[timujinne]

Review summary — two independent reviewers

Reviewed twice, independently, from the same brief: Claude Opus (adversarial, with mutation testing) and GLM-5.2. GLM: APPROVE. Opus: NEEDS-WORK on four small items, no bugs. All four are now fixed in db06137; this comment records what was checked and what was decided.

What was verified, not assumed

• The data is right, against two independent oracles. All 249 shared codes diffed against CLDR 48 et/territories.json → 0 divergences. All 250 values cross-checked against et.wikipedia article titles via the API (redirects followed) → 244/250 resolve exactly; the six that don't are CLDR's descriptor-style forms (Bouvet’ saar, Heardi ja McDonaldi saared, Macau erihalduspiirkond, Myanmar (Birma), Palestiina alad, Pitcairni saared), not misspellings.
• No duplicate keys — all 16 locale files re-parsed with object_pairs_hook, which catches duplicates that a normal JSON load silently collapses. Key set: et − en = {AN}, en − et = {}.
• The capitalisation that looks wrong is correct. Marshalli Saared capitalises the generic noun while Falklandi saared does not — that is the Estonian rule (sovereign state vs. dependent territory), not an inconsistency.
• The 250-vs-249 asymmetry is pre-existing, not introduced here. fr.json already carries CP (Clipperton), which no other locale has and which isn't even in the country data. Translations is the only reader of these files and both public functions are nil-safe.
• Encoding is clean across all 16 files: no BOM, no NBSP, no zero-width characters, all NFC, no mojibake.
• mix test → 121 doctests, 160 tests, 0 failures. mix format --check-formatted → clean. mix credo --strict → no issues.

Fixed after review (db06137)

1. CHANGELOG entry. The convention was reconstructed from this repo's own history rather than assumed: c306bf2 had the contributor write entries under ## Unreleased, and the release commit 8d8635f simply renamed that heading to 1.1.0. So the entry belongs in the feature PR — added, with @version left untouched at 1.1.0. (The "15 languages" line at CHANGELOG.md:246 sits inside the historical 1.0.1 entry and was deliberately left alone — it was true then.)
2. The coverage test was too weak, and this was proven by mutation, not argued. Both predicates tested truthiness, and "" is truthy in Elixir. With et.json["DE"] = "": old predicate → 0 failures; new predicate → 1 failure. The earlier probe was worse still — ZW="", ZM=" ", VN="Estonia???" together produced a fully green suite. Now a nil, empty or whitespace-only name fails on both the English filter and the Estonian reject.
3. dev_docs/2026-03-28_roadmap.md advertised 15 locales; updated to 16 with Estonian checked off. Precedent: c306bf2 updated that file alongside code.
4. AN's spread — see below.

Decisions recorded

• AN stays in et and ru only. The reviewer's point is fair — this turns "uniformly missing" into "present for two locales" — but priv/data/countries/AN.yaml supplies usable names for only five more locales (de, fr, es, ja, nl); the rest would have to be written blind in Arabic, Korean, Chinese. Estonian and Russian are the two this change exists to serve, and both strings were checked against their exact Wikipedia article titles and against each file's own house style (ru.json spells острова out where CLDR abbreviates to о-ва — the hand-written string follows the file, not CLDR). Completing the other locales is a data task of its own.
• Register: this file is 100% CLDR, the existing corpus leans to ISO long names. Measured divergence from CLDR per locale: pl 19, nl 21, it 31, zh 33, de 34, sv 34, en 37, fr 40, es 41, ko 42, ja 47, uk 51, ru 55, pt 63, ar 102 — and et 0. So Estonian says Venemaa where English says Russian Federation and Russian says Российская Федерация. This is deliberate: the strings exist to fill a country dropdown, where Venemaa is what an Estonian speaker expects to read, not Venemaa Föderatsioon. Flagging it explicitly because it is a systematic 37-entry difference rather than a handful of stylistic choices — if the maintainer prefers corpus consistency, CLDR ships alt forms (CZ-alt-variant Tšehhi Vabariik, HK-alt-short Hongkong, PS-alt-short Palestiina, …) and the swap is mechanical.

Left as-is, deliberately

• CD: "Kongo DV" is the only abbreviation in the entire 16-locale corpus, and sits oddly next to this file's own spelled-out CG: "Kongo Vabariik". It is CLDR's current Estonian form; changing it would be the first hand-deviation from the source. Noted for the maintainer.
• BV: "Bouvet’ saar" uses U+2019 where the corpus convention is U+0027 (11 occurrences across en/fr/it/nl/ru vs 2 in uk.json). Also CLDR-verbatim. Real but small impact: naive substring matching or a user typing ' won't match.
• SZ: "Svaasimaa" is not a stale pre-2018 name — et.wikipedia redirects Eswatini → Svaasimaa, so it is prevailing Estonian usage. Recorded so nobody "fixes" it wrongly. CLDR offers SZ-alt-variant: "eSwatini" if cross-locale consistency is preferred.
• NL: "Holland" / AN: "Hollandi Antillid" / BQ: "Kariibi Madalmaad" — the first two agree, BQ is CLDR's outlier (et.wikipedia titles it Kariibi Holland). Inherited, not introduced by the hand-written AN string.

Both reviewers converged on ship once the four items above were addressed.

[timujinne]

Correction to the review summary above

Point 1 of "Fixed after review" is withdrawn: CHANGELOG.md and @version are the maintainer's here, so this branch no longer touches them. The review's reading of the repo history (contributor writes under ## Unreleased, release commit renames the heading) was accurate as a description of past commits, but the project's actual division of labour puts both in the maintainer's hands.

The commit was amended accordingly — the branch now carries only the two code/doc fixes, and the release-note material lives in the commit message instead:

• Estonian (et), the 16th locale for Translations.get_name/2 and get_all_names/1, country names from CLDR (unicode-org/cldr-json, et territories) — verified against CLDR 48 with 0 divergences and cross-checked against et.wikipedia titles.
• AN (Netherlands Antilles) for Estonian and Russian. CLDR dropped the retired code, so no locale file carried it and it surfaced as a lone English row in translated dropdowns — in Russian ahead of every Cyrillic entry, since Latin bytes sort first.

Everything else in the summary stands: the coverage-test fix (proven by mutation) and the roadmap locale count are unchanged.