Restore "Migrate UNIT_AND_LEVEL_TO_LESSON_S3_NAME to ai_rubric_s3_config on unit model"

[davidsbailey]

Background

Reverts #71278, restoring #71016.

I didn't get a complete root cause on the original PR, but the problem was that seeding failed during the seed step during deploy in the test and production chef-managed environments. The issue was able to reach production because I manually worked around the issue in test by explicitly seeding each ai-rubric-enabled unit via cd dashboard; bundle exec rake seed:single_script SCRIPT_NAME=allthethings (for example).

Here is the error log from the initial failure in the test environment (slack):

Aws::S3::Errors::NoSuchKey: Error parsing script file ./config/scripts_json/allthethings.script_json: The specified key does not exist.
...
Caused by:
Aws::S3::Errors::NoSuchKey: The specified key does not exist.
...
/home/ubuntu/test/dashboard/app/jobs/concerns/ai_rubric_config.rb:39:in `read_file_from_s3'
/home/ubuntu/test/dashboard/app/jobs/concerns/ai_rubric_config.rb:89:in `get_s3_learning_goals'
/home/ubuntu/test/dashboard/app/models/learning_goal.rb:29:in `validate_ai_config'
...
/home/ubuntu/test/dashboard/lib/services/script_seed.rb:736:in `import_learning_goals'

I was able to determine that the error occurred because ai_rubric_s3_config did not appear in the properties field on the unit object at the time import_learning_goals was called, even though it should have been set earlier in the same file when the unit was seeded:

• code-dot-org/dashboard/lib/services/script_seed.rb

  Line 225 in 2121317

  seed_context.script = import_script(script_data, md5: md5)

• code-dot-org/dashboard/lib/services/script_seed.rb

  Line 272 in 2121317

  seed_context.learning_goals = import_learning_goals(learning_goals_data, seed_context)

After reverting the original PR, I wasn't able to get a local repro.

Description

Since I wasn't able to get a local repro, but I was able to debug far enough to determine that the problem was ai_rubric_s3_config not getting seeded early enough, the new approach I'm going with here is to have this PR add ai_s3_rubric_config to the unit model and the affected units, and then change the implementation to start using the new config in a next PR (by reverting 8830412).

Caveats

Since a root cause of the initial issue was not determined, there's some risk that adding a new ai-enabled rubric to a new unit could cause similar problems. if that happens, we can investigate at that time.