Translation follow-ups: deleted model shown as cloud, Ollama checks on the UI thread #985

Follow-ups from the review of #836 (offline translation through a local Ollama).

1. A deleted model is reported as a cloud model

TranslationService.CheckModelRemoteCloudAsync refuses on any error, including a 404 from /api/show.

1. Pick a model in Manage Ollama, for example qwen2.5:3b.
2. Remove it with ollama rm qwen2.5:3b.
3. PreferredOllamaModel still names it, so every translation now fails with the cloud-model security message.

Fix: on a 404, refuse with a "model not found" message (a new error.* key in all 19 language files). Keep the security refusal for every other failure.

2. Ollama checks on the UI thread

• TranslatorConverter.PerformCheck() is a synchronous HTTP check of up to 250 ms. It runs on the UI thread from Engines → Recheck, from closing the Manage Ollama window, and from Start / Stop Server.
• Each status change posts catalog.Refresh() to the UI thread, so the next donut or Engines page re-checks every engine there. That's the startup freeze that CheckInBackgroundAsync in App.axaml.cs avoids.

Fix: run PerformCheck through Task.Run, and follow catalog.Refresh() with _ = catalog.CheckInBackgroundAsync();.