feat: optional vision flag in OpenAI LLM config

[JumpLink]

Depends on

LibreChat-AI/agents#257 (the agents-side change that consumes this flag). Draft until that lands - this option is a no-op without it.

Problem

When an agent routes image content to a model with no vision support, OpenAI-compatible providers reject the whole request:

• model is not a multimodal model
• No endpoints found that support image input

This happens in practice when a tool returns image content (e.g. an image-generation tool whose base64 artifact is fed back as context) and the active model is text-only. Today there is no way to tell the OpenAI client "this model can't take images, drop them."

Change

Add an optional vision flag to getOpenAIConfig / getOpenAILLMConfig, forwarded onto the OpenAI llmConfig (OAIClientOptions.vision). When a caller sets vision: false, the chat client strips image content before sending.

• OpenAIConfigOptions.vision?: boolean (public input) and OAIClientOptions.vision?: boolean (the client options).
• getOpenAILLMConfig sets llmConfig.vision when the flag is provided; otherwise it is omitted.
• Scoped to the OpenAI branch only (the ChatOpenAI / Azure / DeepSeek / xAI family that agents#257 covers). Anthropic/Google are untouched.

Defaults to undefined, so existing behavior is unchanged unless a caller opts in.

Consumption

The image stripping itself lives in @librechat/agents (LibreChat-AI/agents#257): ChatOpenAI/AzureChatOpenAI/ChatDeepSeek/ChatXAI read vision from their constructor fields and strip image_url parts before streaming. The llmConfig produced here becomes those constructor fields, so this flag drives that behavior.

The capability source (a model spec / agent vision boolean → options.vision) is intentionally left to the caller; this PR is the config-layer plumbing.

Tests

getOpenAILLMConfig unit tests cover vision true / false / absent.

Testing status

Built and unit-tested at the @librechat/api package level (typecheck clean, 85/85 in llm.spec.ts). Not yet end-to-end integration tested against agents#257, since that change is unpublished - hence draft.