Agent: Refactor the code so that it's not raising exceptions, or producing warn logging messages, but works with `Try`

## Background
We don't want the user's code to fail if the agent is called - even with invalid input. Currently there are some places in agent's code with exceptions being raised - please change it to Success/Failure with Try containers.

Also, let's first discuss with the team whether this approach generally speaking makes sense or whether it would be good to parametrize this somehow, or so.

Obvious note: This ticket will change the data model, at least the internal one on the agent.