Task response formats

[brownsarahm]

need to provide options for strucutred responses

maybe in templates to request json or as an additional bench-level meta-task

ollama CLI has --format json option that seems to work, need to figure out way to use that through python/ how to let users pass that

https://github.com/ollama/ollama-python/blob/main/examples/structured-outputs.py

Still to do:

• make it work for other runners
• decide how to store format type in each task format or if per benchmark?