fix(i18n): drop the retired "shown once" claim from the generate-key toast

[argszero]

Summary

Generating an API key in Settings toasts a claim the product retired in v1.22.1:

已生成新 API Key「{name}」（完整 key 已展示，仅此一次）
New API Key “{name}” generated (full key shown once)

Both sentences were true when they were written (68f9f70, #86, v1.21): the list endpoint returned
only a masked key, the full value came back from POST /api/api-keys and was kept in a session-only
Live.fullKeys cache that a refresh dropped. #95 (799c18d, v1.22.1) closed that state — its own
message records "removed the Live.fullKeys session cache and the 'shown once' logic":

• GET /api/api-keys now returns the owner-visible full_key next to the masked key
  (src/dao.rs; pinned by routes::tests → "撤销前列表返回属主完整 key")
• copyKey() copies k.full_key straight from the list data, with the masked key only as a
  fallback — so the full key is retrievable at any time, including after a reload
• ui/README.md documents it as designed behaviour: 「一键复制完整 key……列表脱敏展示
  atk_live_****xxxx，复制为完整值」

#95 removed that logic and one of its carriers (settings.ak.copy.once, zh/en). This one stayed
behind, and so did a comment in the same function — while the code three lines below copyKey()
still states the policy the string violates:

// 防御性兜底：full_key 缺失时退化为复制脱敏 key，绝不提示「仅生成时展示一次」

So the settings page tells the user to treat the key as unrecoverable while the list keeps a button
that recovers it. The wording is the only thing wrong: the mask-and-copy behaviour is fine and is left
alone.

Changes

• ui/js/i18n.js — settings.ak.gen.ok in both packs, value only:
  • zh: 已生成新 API Key「{name}」（完整 key 可在列表中随时复制）
  • en: New API Key “{name}” generated (full key can be copied from the list anytime)
• ui/js/app.js — the sibling comment that repeated the same retired fact
  (「完整 key 仅生成时可得」) now names the mechanism instead (full_key ships with the row).
• ui/index.html — cache-bust tokens for the two changed assets (i18n.js, app.js → 20260928-1).

No key is added or removed (688 keys before and after; the {name} slot is kept in both packs), no
backend change, no behaviour change. The declared-unreachable settings.ak.gen.ok.mock is left exactly
as it is — it is registered in UNREACHABLE_PACK_KEYS as old-design, and that axis is closed.

No gate. The value restates a retired fact rather than encoding an invariant, and the repository's
standing decision on that axis is to fix the data and not to guard the prose (precedents: #309, #310,
#312–#315).

Related Issue

Tests

• cargo test — 417 passed / 0 failed (the run at the parent commit is the same 417; the
  change touches no tested byte but ui/js/i18n.js is read by the i18n gates, so the count
  matters)
• cargo fmt --check — clean
• cargo clippy --all-targets -- -D warnings — clean
• A/B verification script (tmp/r110/verify.py, run against the tree under test via git -C):
  on the pre-fix tree 8 legs are red — both pack values, "no shipped string value carries the
  fact" (2 of 1701), "every surviving site is a comment line", and the three cache-bust legs —
  on this branch all legs are green. The four legs that assert the behaviour the new sentence
  describes (list returns full_key, copyKey consumes it, the list is reloaded after generating,
  the Rust test pinning it) pass on both trees, which is the point: only the wording moved.

Checklist

• 分支命名符合约定（fix/）
• Commit message 使用 Conventional Commits 格式
• 单一职责，改动最小化

[argszero]

Self-review (committer; allow_self_merge: true is configured for this task).

Scope — 3 files, +7/−5, no production logic touched: two i18n string values (zh + en), one comment in ui/js/app.js, and the two ?v= cache-bust tokens for exactly the two changed .js files.

What the change is — settings.ak.gen.ok claimed "full key shown once" / 「仅此一次」. That fact was retired in 799c18d (#95, v1.22.1), which made the full key copyable at any time from the API-key list. The claim and the product have contradicted each other since. The toast now states the behaviour the code actually has.

Verification

• cargo test on the branch: 417/0; cargo fmt --check rc=0; cargo clippy --all-targets -- -D warnings rc=0.
• CI on this PR: msrv pass (26s), test / fmt / clippy pass (2m17s).
• Repo-internal evidence harness (A/B): an independently materialised tree at the base commit fails the check (the shipped string still carries the retired fact), while the worktree passes all checks.

Blast radius — settings.ak.gen.ok.mock is untouched and remains in the src/i18n_pack.rs sunset list. No new i18n keys are added, so the pack counters (ZH/EN_KEY_COUNT, T_LITERAL_*, STATIC_ATTR_*) are unchanged. ui/index.html carries only the two cache-bust bumps.

No outstanding review feedback; merging.