Implement ollama.auto_select_first from settings

[rosspeili]

Problem Statement

rooms.settings.example.yaml defines ollama.auto_select_first, and OllamaSettings loads it, but nothing in the codebase reads this flag. Users with many local tags still type or hardcode model strings; beginners may not know which tag to use after ollama pull.

Proposed Solution

• When ollama.auto_select_first: true and the effective model is an Ollama/LiteLLM local route, query Ollama’s model list (e.g. http://localhost:11434/api/tags or existing LiteLLM/Ollama helper) and use the first available tag if the configured model is missing or set to a sentinel like ollama/auto.
• Default remains false; built-in default model stays ollama/gemma4:e2b when no file or flag is set.
• Clear CLI message when Ollama is unreachable: fall back to configured defaults.litellm_model with a warning, do not hang the wizard.
• Document behavior, limits, and failure modes in docs/ARCHITECTURE.md / docs/EXAMPLES.md.
• Tests with mocked HTTP (no real Ollama in CI).

Alternatives Considered

• Always pick first Ollama tag on every run — rejected; breaks users with intentional model pins in YAML.
• Remove auto_select_first from the schema — rejected; field is already shipped in the example file; better to implement or deprecate explicitly in a separate change.

Priority

Low 🟢

Additional Context

Design needs explicit UX: only runs when enabled in settings, respects defaults.litellm_model as fallback. Consider interaction with future preset wiring (issue: presets in CLI).