fix: stabilize opencode model selection

[gricha]

Summary

• prefer workspace OpenCode model discovery when zen token is configured, bypassing cached host-only results
• clear model cache on agent config updates so model lists refresh immediately
• display provider/model labels in chat and mobile model pickers to make selections unambiguous
• add coverage for cache selection logic in model discovery tests
• remove the docs deployment workflow to stop the failing pipeline
• resolve lint warnings flagged by CI

Testing

• bun run test:unit
• bun run lint