feat(vision): add auxiliary vision model for image description fallback

[Johell1NS]

What

Adds an auxiliary vision model fallback. When the user sends an image to a model that does not support image input, the bot:

1. Automatically detects the current model lacks vision capability
2. Sends the image to a vision model configured by the user (via /setvision)
3. The vision model describes the image (with dynamic prompt: if a caption is provided, describes in context of the user question)
4. The textual description is forwarded to the main model, which continues processing

Everything is automatic — no user confirmation required. If no vision model is configured, the bot behaves exactly as before (error message + text-only fallback).

Closes #151

Commands

• /setvision — shows a menu to select a vision model. If already set, shows the current model with "Change" and "Clear" options
• The vision model selection persists in settings.json across restarts

Technical changes

3 new files:

• src/app/services/vision-model-service.ts — service: describeImage() creates a temporary session, calls session.prompt() synchronously with the vision model, extracts the text description, deletes the session
• src/bot/commands/setvision-command.ts — /setvision command handler
• src/bot/callbacks/vision-model-callback-handler.ts — inline menu callback for vision model selection

17 modified files:

• src/app/types/settings.ts / src/app/stores/settings-store.ts — new currentVisionModel field
• src/bot/handlers/photo-handler.ts — vision fallback in !supportsImage branch
• src/bot/handlers/document-handler.ts — fallback for image documents
• src/bot/handlers/media-group-handler.ts — fallback for media groups with images
• src/bot/routers/command-router.ts / callback-router.ts — command and callback registration
• src/bot/menus/inline-menu.ts — added "vision" menu kind
• src/bot/commands/definitions.ts — command list registration
• src/i18n/*.ts — 11 new i18n keys across 7 languages

Notes

• The temporary vision session is created as a child of the foreground session to suppress background session notifications
• The vision model call uses session.prompt() (synchronous) rather than promptAsync(), since the full response is needed to continue the flow
• The image description prompt is dynamic: if the user wrote a caption, the prompt is "The user asked: [caption]. Please describe the image focusing on the most relevant aspects." — otherwise "Please describe this image in detail."