docs(ui): stop restating the i18n counters and correct two stale gate claims

[argszero]

Summary

The i18n section of ui/README.md stated counters and guard states the code has moved past, so a
reader today learns numbers the gates no longer report and reads about a missing guard that in fact
exists. Five claims in that section have drifted; each was true when it was written.

• Pack size frozen as "806 keys ×2" — true when 5183e41 (fix(i18n): register every backend error in the wordlist, and stop the comment-stripper from mangling UTF-8 #249) wrote it. The positive-control
  constants in src/i18n_pack.rs own the real value and fail loudly whenever the packs change
  (ZH_KEY_COUNT / EN_KEY_COUNT).
• "431 T() literals", "305 data-i18n*", "681-key union" on the key-consistency line. The
  first two come from 818ed88 (fix(ui): label a market row's availability from the row, not from a missing key count #248) and fcff194 (test(ui): guard the language-pack invariants that were previously unchecked #170); the 681 is owned by nothing — no
  constant in src/i18n_pack.rs, and no smoke-test file (the repository ships only
  api.js / app.js / data.js / i18n.js under ui/js/).
• "Orphan keys are unguarded today (60-odd unreachable keys)" in the nested-data-i18n A/B
  note. every_pack_key_reaches_a_consumer landed two PRs later (a4cb622, test(i18n): every pack key must reach a consumer #266); it asserts the
  computed unreachable set equals UNREACHABLE_PACK_KEYS, so the competing fix m_drop_parent
  is rejected today — it makes the key unreachable while the sunset list stays put. The note also
  contradicted the section directly below it, which documents that very gate.
• "Shrinking the sunset list is a separate follow-up round" in the same section, although
  77b2a82 (chore(i18n): drop 23 keys that no consumer can reach #270) landed it.

Changes

• ui/README.md: drop the restated counts and point at the mechanism that owns them; state that
  the orphan-key axis now has a gate, and that the list shrink has landed.

The counts are not refreshed to today's values. A number a gate already owns is a second carrier
of one fact; restating it here is how this drift happened in the first place, and the constants fail
loudly on their own. This is the treatment 45510ba (#309) gave the perf-gate header, and #180
before it. The "681" is simply removed — nothing carries it. No gate is added for prose: this
repository has already ruled that documentation claims are corrected as data, not fenced in by a new
test.

Related Issue

None — the repository has no open issues, and this is a documentation-consistency fix found by recon,
not a reported bug.

Changes

• 功能/修复说明 — five stale claims in the i18n section corrected; nothing else touched
• 涉及配置/数据结构的改动已同步示例文件 — n/a, docs only

Tests

• cargo test 全部通过 — 409 passed; 0 failed
• cargo fmt --check 通过 — exit code 0
• 新增/更新了单元测试（如适用） — n/a, docs only; the gate that reads ui/README.md
  (state_gate's toast-vocabulary contract line) is untouched and still green

Also ran cargo clippy --all-targets -- -D warnings: exit code 0.

Checklist

• 分支命名符合约定（feat/ /fix/ /docs/ ...） — docs/ui-readme-i18n-claims
• Commit message 使用 Conventional Commits 格式 — docs(ui): …
• 单一职责，改动最小化 — 1 file changed, 7 insertions(+), 5 deletions(-)