Custom provider `providerOptions` should be transparently passed to API request body

## Problem When using a custom-type provider (e.g. ZenMux) with `zenmux/auto` model routing, the `model_routing_config` parameter is required in the `/chat/completions` request body, but Alma's `providerOptions` mechanism does not transparently pass custom parameters to the API request. After setting `providerOptions` on a model via the API (`PUT /api/providers/:id`), the parameters never reach the actual HTTP request sent to the provider. ## Root Cause (from reverse-engineering `app.asar`) In the main process, for `custom` type providers, Alma correctly maps `providerOptions` into the AI SDK's metadata: ```js // Custom → OpenAI mapping const o = "custom" === n ? "openai" : n; Et[o] = { ...Et[o] || {}, ...e }; // e = model's providerOptions ``` However, the `@ai-sdk/openai-compatible` SDK's `getArgs` filters providerOptions through its schema: ```js Object.fromEntries( Object.entries(providerOptions?.[this.providerOptionsName] ?? {}) .filter(([key]) => !Object.keys(schema).includes(key)) ) ``` This suggests the parameters **should** pass through (unknown keys are kept), but in practice they don't reach the request body. ## Use Case ZenMux's intelligent model routing requires `model_routing_config` in the request body: ```json { "model": "zenmux/auto", "model_routing_config": { "available_models": ["sapiens-ai/agnes-2.0-flash", "google/gemini-3.5-flash", "deepseek/deepseek-v4-pro", "openai/gpt-5.4"], "preference": "balanced" } } ``` Without this parameter, the API returns: `Parameter model_routing_config is required when model is zenmux/auto`. ## Suggested Solutions - **Option A (Fix)**: Ensure `providerOptions` set on a model are transparently spread into the chat completions request body for custom-type providers - **Option B (New feature)**: Add an `extraBody` or `bodyTemplate` field to `StoredProviderModel` that gets merged into the API request body, independent of the providerOptions mechanism - **Option C (Workaround)**: Allow setting fixed body parameters at the provider level that are always included in requests ## Environment - Alma version: v0.0.810 - Provider type: custom (ZenMux) - Model: zenmux/auto