fix: require a contextual translation engine

[fly1d]

Why

Automatic mode silently presented the small fallback dictionary as contextual translation, and the sample paragraph had a hard-coded full translation. That made first-use quality look better than arbitrary text and could invalidate customer feedback.

What changed

• require Ollama or a configured cloud API for automatic contextual translation
• keep the offline dictionary as an explicit word-lookup mode
• remove the sample-specific full translation
• add actionable first-use setup guidance on desktop and mobile
• treat empty or whitespace-only API keys as unconfigured in Web and Tauri
• document a 14-day first-customer validation funnel with a CNY 39 founder offer

Validation

• npm run check
• npm run test:smoke
• npm run desktop:check
• direct API checks for auto-without-engine and explicit dictionary behavior
• manual browser review at desktop and 390 x 844 mobile viewport

[fly1d]

Reviewed the complete diff at e4a8313. Automatic mode now requires a real contextual engine in both Web and Tauri; the offline dictionary is only reachable through explicit selection and no longer contains a sample-specific full translation. Empty and whitespace-only API keys are handled consistently, first-use setup behavior is covered in browser smoke tests, API and Rust regression tests cover the provider boundary, and the 14-day validation plan avoids collecting user content or keys. All five required GitHub checks and the full local suite pass. No blocking findings.