Adding translation support for any language, adding mlx backend for s…

[clarkcancode]

This pull request introduces improved backend support for localization and clarifies configuration options for diarization and Docker networking. The most important changes are grouped below by theme.

Localization and translation support across the entire application

• All error and UI messages now use a translation utility function to provide localized strings instead of hardcoded English/Swedish text. This includes error messages for not found resources, queue failures, and UI labels.

Configuration and documentation improvements:

• The .env.example file and README.md have been updated to clarify environment variables for backend selection (BACKEND), diarization models, and Hugging Face token requirements. Comments now explain how to configure OpenRouter and Docker networking for host/container communication.
• The README now describes the supported diarization backends (torch with pyannote.audio or mlx with sortformer), and specifies Hugging Face token requirements only for the torch backend.
• Added support for custom openrouter-compatible endpoints to enable use of local instance of LMStudio, VLLM, llama.cpp, etc.
• Added explicit instructions and troubleshooting tips for Docker networking, including how to set OLLAMA_BASE_URL for host access from containers and general rules for host/container connectivity.
• Added instructions for installing MLX-specific Python dependencies when using the MLX backend for diarization and transcription. This removes the need for the whisper CLI on mac.

These changes make the application more robust for internationalization and easier to configure for different environments and backends.