Add custom proxy support for OpenAI-compatible APIs

[gbladislau]

PR motivation

This removes the model name check for custom proxys, if for example the naming there is different then the ones used in the HF Router. Thus giving more custom-ability for the user.

Summary

Adds a generic custom-proxy/ local model provider for OpenAI-compatible proxy endpoints.

This allows users to run models behind an internal/custom OpenAI-style API, for example:

CUSTOM_PROXY_BASE_URL="https://proxy.example.com/api/v1/gpt/session-id"
CUSTOM_PROXY_API_KEY="<optional>"
ml-intern --model "custom-proxy/Azure AI/gpt-5.5" "your prompt"

Changes

• Adds custom-proxy/ to supported local/custom model prefixes.
• Resolves custom-proxy/ to LiteLLM OpenAI-compatible params.
• Preserves CUSTOM_PROXY_BASE_URL exactly instead of appending /v1.
• Allows spaces in custom-proxy/ model names, e.g. custom-proxy/Azure AI/gpt-5.5.
• Documents custom proxy usage in the README.
• Adds unit coverage for custom proxy parameter resolution and CLI model validation.