refine the benchmark eval UX

[SLR722]

What does this PR do?

Refine the benchmark eval CLI to have a better user experience to run benchmark eval on some standard benchmarks.

The benchmarks need to be defined as resource in the distro template

improvements include:

• user don't need to pass in arbitrary eval-task-config, they only need to pass in the list of benchmarks they'd like to eval, the model id to be evaluated on and the output dir to store the eval results
• output aggregate results to the output file. aggregate results are typically what user care most

Test Plan

spin up a llama stack server with eval benchmarks defined
run llama-stack-client --endpoint xxxx eval run-benchmark "meta-reference-simpleqa" --model_id "meta-llama/Llama-3.1-8B-Instruct" --output_dir "/home/markchen1015/" --num_examples 5

return

what are inside the output file

[yanxi0830]

Thank you!