Function and tool calling

Results are only compared within a matching group, so a model tested one way is never ranked against a model tested another way.

Emit correct schema-valid tool calls

There is no ranking for this yet. EvalRank is still collecting the evidence to rank it.