Support evaluation on local results

Currently evaluation logic would [aggregates](https://github.com/autogluon/autogluon-bench/blob/master/src/autogluon/bench/eval/aggregate/results.py) result files only on S3, which is only possible when running benchmark in AWS mode. User would want to run the logic on local results as well.