Backfill UserAddOnAssignmentVersion table with historical data from UserAddOnAssignment

Summary

Create a data migration to backfill the UserAddOnAssignmentVersion table with historical data from the UserAddOnAssignment table for assignments created before 2024-11-07. This is needed to ensure AI metrics and analytics have complete historical data.

Background

The UserAddOnAssignmentVersion model with papertrail tracking was introduced on 2024-11-07 via !169180 (merged). However, this means that add-on assignments created before this date don't have corresponding version records, which affects:

1. AI Impact Analytics that rely on historical assignment data
2. ClickHouse synchronization for historical metrics
3. Accurate reporting of add-on usage over time

As discussed in #498633 (comment 2595465956), we need to backfill the _versions table for assignments that don't have any version records.

Proposal

1. Create a data migration that:

   • Identifies UserAddOnAssignment records without corresponding UserAddOnAssignmentVersion entries
   • Creates UserAddOnAssignmentVersion records with:
     • event = 'created'
     • Proper created_at timestamp from the original assignment
     • All necessary fields to maintain data integrity

2. Ensure proper synchronization:

   • Verify that the backfilled records are properly synced to ClickHouse
   • Update sync cursors if necessary to ensure complete data transfer

3. Data validation:

   • Compare ClickHouse data with PostgreSQL data for add-on assignments under gitlab-org
   • Ensure data accuracy and completeness after backfill

4. Documentation:

   • Document the backfill process and any limitations
   • Update AI Impact Analytics documentation if needed

Acceptance Criteria

• Data migration successfully creates UserAddOnAssignmentVersion records for all UserAddOnAssignment records created before 2024-11-07
• Backfilled records have correct created_at timestamps and event='created'
• All backfilled data is properly synchronized to ClickHouse
• AI metrics show consistent historical data after backfill
• Data validation confirms accuracy between PG and ClickHouse
• Documentation is updated to reflect the backfill process

Technical Notes

• Reference query for estimating records: https://postgres.ai/console/gitlab/gitlab-production-main/sessions/41063/commands/126208
• The backfill should preserve the original created_at timestamps from the UserAddOnAssignment table
• Ensure the migration is safe for large datasets and can be run in production

Related Issues

• Parent issue: #498633 (closed)
• Related to ClickHouse sync issues and AI metrics accuracy