Mistral-Small-24B-Instruct-2501 tasks.yml files

[derekk-nm]

SUMMARY:
adds task.yml config files for the models derived from mistralai/Mistral-Small-24B-Instruct-2501.
The task metric values are from the model card for the specific model (though the values for the "parent" come from one of the RedHatAI models, where the evaluation accuracy table shows the measured parent metrics).

[RedHatAI/Mistral-Small-24B-Instruct-2501-quantized.w8a8](https://huggingface.co/RedHatAI/Mistral-Small-24B-Instruct-2501-quantized.w8a8) is currently (202504210830) missing model card content.

TEST PLAN:
All runs are showing failures. these are likely due to the fact that either the server or client do not have configuration settings set exactly as described in the model card's evaluation section.

• mistralai/Mistral-Small-24B-Instruct-2501: https://github.com/neuralmagic/nm-cicd/actions/runs/14573848346 shows an error in lm-eval and errors accessing the completions api endpoint.
• mistralai/Mistral-Small-24B-Instruct-2501-quantized.w4a16: accuracy ... follow up with larger rtol accuracy
• mistralai/Mistral-Small-24B-Instruct-2501-FP8-Dynamic: accuracy ... follow up with larger rtol accuracy success! and with mistral server settings accuracy

[derekk-nm]

a different source has been chose for the ground truth. This PR is now deprecated.