Persist model selection and improve vision-model handling

[LostOne1000]

Priority

Medium

Summary

The plug-in selects the first model returned by Ollama:

props.selectedModel = availableModels[1]

Ollama does not guarantee that the first returned model is vision-capable or appropriate for image metadata generation.

The user’s model selection is also discarded whenever the dialog is reopened.

Expected behavior

The plug-in should remember the selected model and avoid silently defaulting to an arbitrary text-only model.

Implementation notes

• Store the selected model in Lightroom plug-in preferences.
• When loading:
  1. Retrieve the available Ollama models.
  2. Select the stored preference when it still exists.
  3. Otherwise prefer the configured default model.
  4. Otherwise require explicit selection or use a documented fallback.
• Persist the selection when it changes or when the dialog is saved.
• Consider adding a user-configured vision-model preference rather than attempting unreliable model-name inference.
• Display a clear error when the selected model rejects image input.

Acceptance criteria

• The selected Ollama model persists between Lightroom sessions.
• Reopening the dialog restores the previously selected model when available.
• The first arbitrary model returned by /api/tags is not automatically preferred over a valid saved selection.
• A missing saved model falls back predictably.
• Image-input rejection produces a clear recoverable error.
• Existing model discovery behavior continues to work when Ollama is available.