[BIG FEATURE] Add image recognition fallback input flow

[musnows]

Summary

• add an opt-in image recognition fallback for text-only chat models, with persisted settings and an image input settings section
• pre-recognize attached images before sending to text-only models, then append path, MIME, dimensions, and recognized content to the user prompt without base64 or RecognitionModel
• add a gated built-in recognize_image tool for local paths and URLs, plus runtime progress events shown in the chat UI

Root Cause / Behavior

Text-only models previously either rejected image attachments or relied on generic attachment fallback. This change lets users explicitly configure a multimodal recognition model so new eligible conversations can still attach or paste images while preserving historical conversation behavior by comparing the persisted enable time against the first user message time.

Validation

• npm --prefix kun test -- tests/attachment-store.test.ts src/adapters/tool/image-recognition-tool-provider.test.ts
• npm test -- src/renderer/src/lib/attachment-upload-availability.test.ts src/shared/app-settings.test.ts src/renderer/src/agent/kun-mapper.test.ts src/renderer/src/components/chat/MessageTimeline.tool-summary.test.ts src/renderer/src/components/settings-section-image-input.test.ts src/main/ipc/app-ipc-schemas.test.ts
• npm --prefix kun run typecheck
• npm run typecheck

[musnows]

新增独立的多模态模型配置项，支持让多模态模型预处理图片后发给不支持多模态的模型。同时新增识图mcp工具，支持非多模态模型调用工具阅读图片内容。