Add benchmarking workflow to evaluate classification and OD runtimes

[ntlind]

Use Cases

• We want to look at recent commits to figure out which one caused a change in performance
• We want to quickly test performance of potential changes locally using standardized benchmarks

Changes

• Delete stress-testing.yml and associated files since we never trigger these manual jobs
• Create an automatic benchmarking job to track Valor performance for classification and object detection evaluations. The workflow will fail if any of the individual evaluation tests takes longer than 30 seconds.

Output

stale - need to review with Eric.

[ekorman]

we can host the data in a public s3. prefer not having in version control

[ekorman]

we can host the data in a public s3. prefer not having in version control
files:

https://pub-fae71003f78140bdaedf32a7c8d331d2.r2.dev/classification_data.json
https://pub-fae71003f78140bdaedf32a7c8d331d2.r2.dev/detection_data.json

[ekorman]

can we make the benchmark-evaluations job fail if it takes too long?