F7: AI Proxy Resource Card — MVP

## Summary
Standalone, connectable AI Proxy resource card. Provisioned once, wired to containers/services via canvas connections. Users can freely bind/unbind. Removes manual key-acquisition and baseURL setup for AI-agent developers.

## Acceptance Criteria
- [ ] Standalone AI Proxy resource card in canvas
- [ ] Provision once, connect to multiple containers/services
- [ ] Bind/unbind connections freely
- [ ] Zero-setup model access — no manual key acquisition or baseURL config
- [ ] Multiple model backends supported
- [ ] Domestic model backend option included at launch

## Sprint Assignment
- Sprint 1: Primary release — ships to production

## Success Metrics
- Adoption rate among AI-agent-persona users
- Connections per proxy card (reuse rate)
- Token cost savings where domestic models selected

Start by mapping the canvas resource-card entry point, the provisioning flow, and the connection paths for containers and services. Then identify how model backends, including the domestic option, are represented. Done means the card can be provisioned once, freely bound or unbound across multiple services, and provide zero-setup access to the supported models.

Written by the indexing model from the issue text.