Store tool result internally as a Result

[tidely]

The LLM always gets returned a string, which is my assumption why currently all internal representations of a tool call result are simply a string.

The motivation for the change in the title, would be that hooks which make use of a tool result callback, could differentiate between a successful and a failed tool call. This could be rather hard with tools which define a custom Error type, which the callback can't be generic over. But even then we could store the tool call as Result<String, String>, such that the information of failure would be preserved.

Is this something we'd want to have? Are there design issues which would block this?