feat(onboarding): Attach images in onboarding chat for vision-based understanding

[TemaDeveloper]

Size: L — headline onboarding improvement.

Why

Onboarding is a pure text interview (ONBOARDING_SYSTEM_PROMPT, conversational-onboarding.tsx — the composer is a single text input, no attach affordance). People can describe their life far faster by showing it: a photo of their paper planner, a screenshot of a budgeting spreadsheet, a gym program, a class timetable, a meds list, a whiteboard of goals. Letting them attach images makes the derived life-profile richer and the generated sections fit better on the first try.

Scope

1. Client (src/components/onboarding/conversational-onboarding.tsx): add an attach button + thumbnail previews + remove; allow paste/drag-drop; send image refs alongside the text turn.
2. Upload: reuse the existing private-Blob path (/api/notes/upload: image-only, 10MB, ownership-scoped serving via /api/notes/blob) — or a parallel onboarding/<userId>/... namespace. Don't send raw base64 through the chat body.
3. API (src/app/api/profile/chat/route.ts): accept images on the turn; pass them to the vision-capable extraction + reply calls (depends on multimodal callAI — see the multimodal-callAI issue).
4. Extraction (src/lib/profile/facet-extract.ts): feed images so facets are derived from what's shown, not just typed.
5. Update ONBOARDING_SYSTEM_PROMPT to invite "type or show me" and to reason over attached images.

Notes

• Vision inference on multiple images is slow — respect/raise maxDuration (currently 60s) and stream or show progress.
• Privacy: onboarding images can be sensitive (finances, health) — keep them private-access, scoped to the user, and consider deleting them after profile extraction.

Depends on: multimodal callAI. Pairs with: seed-from-artifact, confirm-understanding.

[TemaDeveloper]

Onboarding image-attachments + better-understanding epic — related issues:

• feat(ai): Multimodal callAI — image content blocks for Claude/Gemini/OpenAI/Mistral #64 — Multimodal callAI (enabling infra, do first)
• feat(onboarding): Attach images in onboarding chat for vision-based understanding #65 — Attach images in onboarding chat (this issue — headline feature)
• feat(onboarding): Seed facets, sections & rows from an uploaded artifact (spreadsheet/planner/schedule photo) #66 — Seed facets/sections/rows from an uploaded artifact (vision→structured import)
• feat(onboarding): Confirm-and-refine derived understanding before generating the planner #67 — Confirm-and-refine derived understanding before generating

Suggested order: #64 → #65 → #67 → #66.