Add OpenRouter transcription support

[snavatta]

Summary

• add a first-class OpenRouter transcription engine with server-side authentication and live transcription-model discovery
• send OpenAI-compatible multipart transcription requests, preserving segment timestamps when supported and falling back to standard JSON for other providers
• expose OpenRouter in project/default settings while keeping hosted models out of the local Model Manager
• disclose remote audio processing in the settings UI and document configuration across the README, PRDs, and shared API contract
• run browser tests against a freshly built production preview instead of a reusable development server

Configuration

Set Transcription__OpenRouter__ApiKey on the server. The optional Transcription__OpenRouter__BaseUrl override must be an absolute HTTPS URL.

Verification

• dotnet test src/ClassTranscriber.slnx --no-restore — 217 passed after merging current master
• npm test -- --run — 31 passed
• npm run lint — passed
• npm run test:e2e — 5 passed against the production preview
• live /api/settings/options check — OpenRouter returned 19 current transcription models
• live /api/settings/models check — 0 OpenRouter entries, confirming hosted models remain outside local model management
• responsive settings review at 375px, 768px, and 1280px — passed after adding a discoverable mobile Model Manager scroll affordance