Refactor Human-Machine-Interface

[maciejmajek]

Is your feature request related to a problem? Please describe.
The current HMI implementation is unnecessarily complicated, with vastly different solutions for text and voice interfaces.

Describe the solution you'd like
The HMI should implement a base class for any I/O modality. It's well known that different modalities vary in their ability to convey different aspects of a conversation. For example, a Streamlit/chat-like interface can easily display a thought process, including showing images, while voice communications cannot. The implementation should take these issues into account.

One possible solution would be to use a callback system that attaches to different parts of the HMI agent.