Refactor "In-House" Provider to "Self-Hosted" & Fix API Parameters

[amitb0ra]

Description

The current implementation for certain AI model providers is rigid and prone to failure due to non-standard parameter naming. This issue proposes standardizing the API request structure and refactoring the "In House" provider into a more flexible "Self-Hosted" configuration that supports custom endpoints and private model instances (e.g., Ollama, vLLM).

Current Behavior

• Incorrect Payload Keys: Both RCInHouseHostedModels.ts and GroqModels.ts send non-standard keys like rcModel or groqModel in the JSON body, leading to 400 Bad Request errors from standard OpenAI-compatible APIs.
• Hardcoded URLs: The "In House" provider forces a URL structure of http://${rcModel}/v1, which prevents the use of HTTPS, custom ports, or specific sub-paths.

Proposed Changes

• Parameter Standardization: Update request bodies to use the standard model key across all providers to ensure compatibility.
• Provider Renaming: Refactor the "In House" provider labels across the UI and settings to "Self-Hosted" for better clarity.
• Custom Endpoint and Model Support: Introduce a new setting ID_SELF_HOSTED_URL and ID_SELF_HOSTED_MODEL to allow users to input their own base API path (e.g., https://ollama.internal.net/v1) and model (e.g., openai/gpt-oss-20b)