Evaluations cloud run jobs runner

[warmbowski]

Setup of evaluations application used for maintaining a configuration of model evaluations by "tier" (eval runs based on frequency). The code builds and deploys a docker file to Google Cloud Run to run the evaluations on schedule. The only dependency is the olmo-eval-internal cli tool for running evaluations. See the Readme.md for details.

• Docker image can be built and evals run locally with the run_local.py utility
• Docker image can be deployed manually using typical docker push command.
• Jobs and scheduling can be deployed manually using terraform apply
• Skiff2 is leveraged running setup/build/deploy on push to main.
• ad-hoc evals can be executed with gcloud command line passing --updated-env-vars

TODOs:

• switch setup action to shared GitHub actions when publicly available
• switch build action to shared GitHub actions when publicly available AND olmo-eval-internal is publicly available
  • Remove GH Token access to olmo-eval-internal
• add in storage configurations when shared database is accessible from GCP

closes https://github.com/allenai/playground-issues-repo/issues/994
closes https://github.com/allenai/playground-issues-repo/issues/990