Fix AI provider and transcription setup UX

[SmetankaKluss]

What changed

• adds a no-completion-token connection test for OpenAI and xAI keys/models
• makes the active AI provider explicit and keeps authentication failures inside AI Editor
• disables transcription engines that are not included in the desktop build
• explains that Faster Whisper models download automatically on first use
• bumps the desktop release version to 0.1.2 and updates troubleshooting guidance

Root cause

The settings UI treated any non-empty API-key field as verified and did not explain that only one provider is active. The transcription selector also exposed optional engines such as WhisperX even when their Python dependency was absent from the packaged app.

Validation

• frontend lint and production build
• all frontend smoke workflows
• 41 backend smoke tests
• packaging configuration smoke test
• browser QA of disabled WhisperX, active Grok, connection testing, and friendly topic-edit failure