Extract shared Ollama /api/tags helper (DRY preflight + auto-select)

[rosspeili]

Problem Statement

Ollama tag discovery is implemented inline in rooms/ollama_preflight.py. PR #44 and issue #30 add a second urlopen + JSON parse path in settings.py. Tag matching rules (name vs model, :latest) should live in one place.

Proposed Solution

1. Add e.g. rooms/ollama_client.py (or extend ollama_preflight.py) with:

   • fetch_ollama_tags(base_url, timeout=3.0) -> list[str]
   • tag_is_available(configured_tag, available_tags) -> bool

2. Refactor run_ollama_preflight() to use the helper.

3. When feat: support ollama.auto_select_first model resolution #44/Implement ollama.auto_select_first from settings #30 land, resolve_ollama_model() uses the same helper.

4. Unit tests for tag matching edge cases (exact, :latest, empty list, unreachable server).

Acceptance criteria

• One implementation of /api/tags fetch and tag normalization
• Existing preflight tests still pass
• No behavior change for default preflight path unless bugs found during refactor

Alternatives Considered

• Inline duplication in feat: support ollama.auto_select_first model resolution #44 — rejected; guaranteed drift.
• LiteLLM model list API — rejected; preflight intentionally uses local Ollama HTTP.

Priority

Medium 🟡

Additional Context

No response

[Chirag04-bit]

Hi @rosspeili,

I saw this issue and proactively tackled it directly inside PR #44!

I centralized the /api/tags HTTP request logic and normalization rules to keep it strictly DRY across both the preflight check and the auto-select paths. The existing preflight tests are fully updated and passing along with the new edge cases.

Once you take a look at PR #44, this can likely be closed out alongside it.