feat: Skip provider prefixing in model names if model already contains a slash '/'

[worldwidehigh]

Feature Title

Skip auto-prefixing on models with existing slashes

Describe the feature

Currently, when configuring an OpenAI-compatible provider (e.g., EDGEQUAKE_LLM_PROVIDER=openai), the backend automatically prepends the provider name to the model string (e.g., transforming gpt-4o-mini into openai/gpt-4o-mini).

We propose adding a simple check in the model factory:
If the configured model name already contains a slash / (for example, openai/gpt-4o-mini or deepinfra/minimax-m2.5), EdgeQuake should skip prepending the provider name and pass the model string exactly as configured by the user.

A potential fix in factory.rs could look like this:

let final_model_name = if model_name.contains('/') {
    model_name.to_string()
} else {
    format!("{}/{}", provider_name, model_name)
};

Motivation

When routing API requests through intermediate AI Gateways, Routers, or Proxies (such as Requesty, Portkey, or custom corporate proxies) that utilize the standard OpenAI-compatible client under the hood but require the standard provider/model format, the current auto-prefixing logic results in a double-prefixed string (e.g. converting openai/gpt-4o-mini into openai/openai/gpt-4o-mini).

This causes the upstream router to reject the request with Invalid model, expected: "provider/model" errors.

Allowing explicit model naming by bypassing the auto-prefixer when a slash is already present would make EdgeQuake highly compatible with any custom AI routers and third-party API gateways out of the box, without breaking backwards compatibility.

Alternatives considered

1. Using the openrouter provider as a workaround: While this works for the LLM because it doesn't auto-prepend, it currently disables/mocks embedding endpoints or causes warning loops, making it an unstable hack for fully production-ready RAG setups.
2. Patching the source files (factory.rs, openai.rs) locally: This is possible but maintaining a custom fork diverges from upstream. A native solution is highly beneficial for the community.

Additional context

This issue was discovered while setting up a local EdgeQuake pipeline routed through an EU-compliant Requesty gateway. Once the double-prefixing issue was bypassed, the Rust backend worked flawlessly and yielded incredible performance.