No model picker during `ucode configure` — defaults to first discovered model

[zjacobson1016]

Summary

ucode configure auto-selects the first/default model it discovers for each coding agent and does not prompt the user to choose a preferred model. This makes it difficult to use a specific model version (e.g. databricks-claude-opus-4-6 instead of databricks-claude-opus-4-7).

Expected behavior

During ucode configure, users should be able to pick which AI Gateway model to use per agent (similar to how agent selection works today).

Actual behavior

• ucode configure discovers available models from AI Gateway
• Each agent's default_model() picks the first match by family (e.g. latest opus by reverse sort)
• That model is written into agent config and used on every launch
• No interactive model selection step exists in the configure flow

Workaround (clunky)

For Claude Code, users can override at launch time:

ucode claude -- --model databricks-claude-opus-4-6

Or change model mid-session with /model in Claude Code. Other agents have similar limitations.

Impact

• Users cannot easily pin a specific model version across agents
• Defaults may change when new models appear on the workspace
• Overrides require remembering CLI flags or in-session commands on every launch

Suggested improvement

• Add a model picker step to ucode configure (per agent or globally)
• Persist selected models in ~/.ucode/state.json
• Optionally support model selection via ucode configure --agent claude with a --model flag

Environment

• ucode (latest)
• Databricks workspace with multiple models per family (e.g. opus-4-6 and opus-4-7)