Add multi-LLM fallback support (Pixtral / Mistral / Groq) and UI settings

[Lzxpan]

Motivation

• Add support for multiple LLM providers and automatic fallback so OCR translations can continue when one provider is rate-limited or unavailable.
• Expose provider-specific API keys and model selections in the UI and persist them in user settings.
• Provide OpenAI-compatible clients to reach non-Gemini providers (including local vLLM endpoints) and unify error/retry handling.
• Improve UX on window close by prompting to exit or minimize to tray.

Description

• Added new user settings and defaults for ApiKey_Pixtral, ModelName_Pixtral, ApiKey_MistralPixtral, ModelName_MistralPixtral, ApiKey_Llama4, and ModelName_Llama4 and updated App.config, Settings.settings, and Settings.Designer.cs to persist them.
• Extended Form1 to add model combo boxes and API key textboxes for Pixtral / Mistral / Llama4, initialize option lists, update link labels, and read/save those settings on load/close.
• Reworked close handling in Form1_FormClosing to prompt the user to exit, minimize to tray, or cancel.
• Introduced a fallback orchestration in Form1 using BuildLlmCredentials(), TranslateTextWithFallbackAsync(), and TranslateImageWithFallbackAsync() to attempt multiple configured providers in order and switch on quota/rate-limit or retryable network errors.
• Added LlmClients.cs implementing LlmProvider, LlmCredential, LlmErrorPolicy, and OpenAiCompatibleClient with helpers to call OpenAI-compatible chat endpoints and to post image data as data-URI for OCR+translation.
• Updated HandleCapturedImage to call the new fallback methods and update UI/token usage handling accordingly.
• Included the new LlmClients.cs in the project file (ScreenOCRTranslator.csproj).

Testing

• No automated tests were executed for this change.

---

Codex Task

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 3857a1a79a

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Skip local Pixtral fallback unless user configured it

BuildLlmCredentials() currently adds the local Pixtral endpoint whenever cmbModel_Pixtral.SelectedItem is set, and Form1_Load always selects an item by default, so an unconfigured http://127.0.0.1:8000/v1 provider is injected for all users. In the common setup without a local vLLM server, this implicit step fails before later providers are reached, which can prevent configured Mistral/Groq fallbacks from being attempted after a Gemini quota/rate-limit event.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Let network errors bubble so fallback can try next provider

PostChatCompletion() swallows transport exceptions and returns a synthetic GeminiResult (HttpStatus = 0) instead of throwing. The fallback orchestration in TranslateTextWithFallbackAsync()/TranslateImageWithFallbackAsync() only switches providers for network failures in their catch (Exception) branch, so these wrapped errors are treated as terminal non-quota failures and the chain stops early. A transient connection failure on one OpenAI-compatible provider can therefore block trying later configured providers.

Useful? React with 👍 / 👎.