fix(engines): 候补模型顶替时的实时预览状态与隐私说明

[sunerpy]

Summary

Follow-up to #89 (fallback models, v0.0.35), from the completion review of that work.

• Status follows the stand-in (ResolvedEngines::status_with): the pages now compute the live preview and the output mode from the first model with quota left, in the order the engine's chain asks them (open_order, shared with the chain). When a model that takes whole recordings only stands in for a used-up realtime model, the takes already ran whole; the pages still showed 边说边识别 as in effect. Now live_preview_ready is false, live_source stays stream, effective_output_mode is whole_take; 实时预览 shows "暂不可用 · 候补模型不支持实时识别" and the output mode block no longer asks for a model download. The TS mirror and the mock follow the same rule.
• Privacy line: it named the fallback providers whatever the switch said. It now names them only while the chain runs (switch on, selected service usable).
• Phone: the 语音模型 and AI 模型 pages show the 当前 line (with （候补） while a fallback model stands in) and the privacy line under the provider cards, as the desktop does. Both lines are shared components now (CurrentService, ServicePrivacy in packages/ui); shortModel moved to @voltip/shared.
• Flaky test: one_request_at_a_time_and_a_preview_of_a_closed_sentence_is_dropped (from feat(dictation): the built-in service previews while you speak #77) read the fake's call count before the request's task had started: 16 of 640 runs failed with every core busy. It now waits for the call; only the decode thread's feed/poll send requests, so the count cannot move meanwhile.
• Docs: dictation.md §3.5, frontend.md, acceptance.md, site cloud and privacy pages in both languages.

Testing

• make verify on the pushed head 7b9cd66 (overall_exit=0, log bound to HEAD).
• New regression tests, each failing without its fix: regression_the_status_shows_the_mode_the_next_take_runs_with (engine vs status), the_status_follows_the_model_that_stands_in_for_a_realtime_one, the TS mirror test, ServiceSummary.test.tsx, two desktop page tests, one phone page test.
• With every core busy (32 busy loops next to 32 workers): the fallback engine, chain and redecode tests 640 runs, tests/fallback.rs 320 runs, the long-take fallback test 320 runs — no failures on 7b9cd66 (before the test fix: 16 of 640 failed, all in the redecode test above).

🤖 Generated with Claude Code

[codecov]

Codecov Report

❌ Patch coverage is 96.21212% with 5 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
crates/voltip-core/src/dictation/redecode.rs	50.00%	2 Missing ⚠️
...ackages/ui/src/features/engines/ServiceSummary.tsx	93.10%	0 Missing and 2 partials ⚠️
packages/shared/src/engine-drafts.ts	50.00%	0 Missing and 1 partial ⚠️

@@           Coverage Diff           @@
##             main      #94   +/-   ##
=======================================
  Coverage   93.48%   93.48%           
=======================================
  Files         335      336    +1     
  Lines       47163    47250   +87     
  Branches     3143     3150    +7     
=======================================
+ Hits        44088    44173   +85     
- Misses       2669     2672    +3     
+ Partials      406      405    -1     

Flag	Coverage Δ
rust	93.48% <97.56%> (+<0.01%)	⬆️
web	93.51% <94.00%> (+0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
...desktop/src/pages/settings/engines/EnginesPane.tsx	100.00% <ø> (+2.98%)	⬆️
...desktop/src/pages/settings/engines/LivePreview.tsx	100.00% <ø> (ø)
.../desktop/src/pages/settings/engines/OutputMode.tsx	100.00% <100.00%> (ø)
apps/desktop/src/pages/settings/engines/helpers.ts	100.00% <100.00%> (ø)
apps/desktop/src/shell/page-meta.ts	95.94% <ø> (+1.20%)	⬆️
apps/mobile/src/screens/AiModels.tsx	84.61% <ø> (ø)
apps/mobile/src/screens/SpeechModels.tsx	88.23% <ø> (ø)
crates/voltip-core/src/dictation/engine.rs	97.36% <100.00%> (-0.06%)	⬇️
crates/voltip-core/src/dictation/fallback.rs	98.89% <100.00%> (+0.01%)	⬆️
crates/voltip-core/src/dictation/mod.rs	100.00% <ø> (ø)
... and 8 more

... and 2 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.