Drop legacy bucket_key/requested_bucket_key columns

[nbudin]

Summary

Drops the legacy bucket_key/requested_bucket_key string columns (and the bucket_key_null_for_non_slot_occupying_states check constraint on signups) that #11871 deliberately kept in place for one release, so a rolling/partial deploy couldn't have old code hit a missing-column error while new code was already writing bucket_id/requested_bucket_id.

• signups: drops bucket_key, requested_bucket_key, and the now-redundant check constraint (superseded by bucket_id_null_for_non_slot_occupying_states from Convert bucket_key/requested_bucket_key to real foreign keys #11871).
• signup_requests, signup_ranked_choices: drop requested_bucket_key.
• signup_changes (audit log): drops bucket_key, requested_bucket_key — the bucket_name/requested_bucket_name snapshot columns added in Convert bucket_key/requested_bucket_key to real foreign keys #11871 already cover display needs after a bucket is deleted.
• Migration is reversible (re-adds the columns and check constraint on rollback, though not the historical data).

Verification before merging

Since this drop is only safe once the #11871 backfill is confirmed correct and the new code is actually live:

• Pulled a fresh production database dump (after the Convert bucket_key/requested_bucket_key to real foreign keys #11871 deploy had been running for a while) into development and ran a verification script against it, checking all four tables for backfill gaps, id/key mismatches, and unresolvable orphans:
  • Zero unbackfilled rows outside the documented pre-existing-orphan case (buckets removed from a policy before the migration ever ran) — orphan rate well under 1% on every table, matching Convert bucket_key/requested_bucket_key to real foreign keys #11871's stated known limitation.
  • Zero id/key mismatches anywhere.
  • Confirmed live post-deploy signups have bucket_id set with no legacy key ever written (the new code path), with none of that pattern appearing on rows from before the deploy timestamp — i.e., the cutover is real and clean, not just a coincidence of the backfill.
• Applied and tested the migration itself locally: migrates and rolls back cleanly against the current schema, relevant model/service/GraphQL/notifier/presenter tests (125 tests) pass against the new schema, and confirmed no application code still references the raw legacy columns (only the computed bucket_key/requested_bucket_key reader methods and the unrelated registration_policy_buckets.key column, both of which are intentionally kept).

Fixes #11872

🤖 Generated with Claude Code

[github-actions]

Code Coverage Report: Only Changed Files listed

Package	Base Coverage	New Coverage	Difference
app/javascript/setupI18Next.ts	🟠 65.52%	🟠 72.41%	🟢 6.89%
Overall Coverage	🟢 55.88%	🟢 55.89%	🟢 0.01%

Minimum allowed coverage is 0%, this run produced 55.89%