Add multi-LLM clients, quota tracking and quota-board UI

[Lzxpan]

Motivation

• Add support for multiple LLM backends (local vLLM Pixtral, Mistral Pixtral, Groq Llama4 and Gemini) and allow automatic fallback between configured credentials.
• Track and persist daily usage and simple rate limits so users can view per-engine consumption and identify quota/rate-limit failures.

Description

• Introduce LlmClients.cs with LlmProvider, LlmCredential, OpenAiCompatibleClient, and LlmErrorPolicy to call OpenAI-compatible endpoints and unify text/image requests.
• Add DailyQuotaTracker.cs to persist daily usage to usage_daily.json, provide defaults for common models, and expose snapshot/recording APIs.
• Extend the main form (Form1.cs and Form1.Designer.cs) to add API key/model fields for Pixtral, Mistral Pixtral and Llama4, implement BuildLlmCredentials(), fallback methods TranslateTextWithFallbackAsync() and TranslateImageWithFallbackAsync(), and wire quota recording and saving.
• Add a simple QuotaBoardForm embedded in Form1.cs to visualize today's per-engine usage and a toolbar button btnQuotaBoard to open it.
• Persist new settings by adding properties in Properties/Settings.* and update App.config defaults for the new keys.
• Update project file to include the new source files (LlmClients.cs, DailyQuotaTracker.cs) and include UI/behaviour changes like a close-confirmation dialog.

Testing

• No automated tests were added or executed for this change.

---

Codex Task

[chatgpt-codex-connector]

💡 Codex Review

ScreenOCRTranslator/Form1.cs

Lines 877 to 878 in bc5f9af

	if (cmbModel_Pixtral.SelectedItem != null)
	{

Skip local Pixtral credential unless explicitly configured

BuildLlmCredentials() unconditionally adds the local vLLM Pixtral endpoint whenever cmbModel_Pixtral has a selected item, and Form1_Load always selects one by default. That means users who only configured cloud keys still get http://127.0.0.1:8000/v1 inserted into the fallback chain; when that local server is absent, Translate*WithFallbackAsync() returns this non-quota failure immediately and never reaches later configured providers. Gate this credential behind explicit user intent/configuration so fallback can reach actually configured backends.

---

ScreenOCRTranslator/LlmClients.cs

Lines 141 to 145 in bc5f9af

	catch (Exception ex)
	{
	return new GeminiResult
	{
	HttpStatus = 0,

Let transport errors trigger fallback to the next credential

OpenAiCompatibleClient.PostChatCompletion() catches network/transport exceptions and returns a normal GeminiResult (HttpStatus = 0) instead of throwing. In the fallback methods, retry-to-next-provider on network problems is only implemented in the exception path (catch + IsRetryableNetworkError), so these wrapped failures are treated as terminal errors and stop fallback. This breaks multi-provider resilience during transient connection issues; either rethrow retryable exceptions here or classify HttpStatus == 0 as retryable in fallback.

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".