Refactor how we process eval dependencies

[jb3]

Right now, our eval dependencies are pulled in using a script embedded within the deployment file, and all dependencies are listed there. This gets a bit messy the more dependencies we want to add, especially if we want to add version markers.

This PR splits out the dependencies into a new file under requirements/ and moves the dependency install process to it's own init container instead of being a task that executes within the snekbox container after it starts. This also allows for greater observability into the dependencies pull process as logs can be fetched from init containers with kubectl logs deploy/snekbox -c deps-install.

[jchristgit]

Great Dane

[coveralls]

coverage: 90.947%. remained the same when pulling 4251c08e1b2797e4e1d348d2bce13eb1e7bd4051 on jb3/refactor-eval-deps into 02926616c635950a6b2b8cc25277f3cc5bcea131 on main.

[jb3]

Thanks!