[AI Evaluation] Quality evaluators should exclude tool call messages by default #6043
Description
See discussion here. Currently, all the evaluators included in the Microsoft.Extensions.AI.Evaluation.Quality package will render tool call messages as strings with prefix ToolCall and empty content.
Ideally, tool call messages should be excluded from the rendering by default since the Quality evaluators are designed to assess the quality / efficacy of the LLM responses in the user visible portion of the chat conversation. At the same time, the base classes in the Quality package can be used to author other (custom) derived evaluators that may want to take tool calls into account - so we should leave the door open for derived implementations to override the default rendering of messages.
FYI @peterwald