fix(genesis): carry the entity fields the writer never put on Create

[rickyrombo]

Three groups of source columns never reached the chain because genesis-writer did not select them, or because the field only exists on an action the migration never emits.

Tracks and playlists — DDEX rights metadata

The writer selected neither the contributor lists nor the copyright and rights-controller records, so they could not be indexed regardless of what the indexer reads. Added to the Create metadata for both:

tracks	playlists
artists	artists
resource_contributors	ddex_release_ids
indirect_resource_contributors	copyright_line
rights_controller	producer_copyright_line
copyright_line
producer_copyright_line
parental_warning_type

This is live metadata, not a legacy shape: ddex-processor/src/publishRelease.ts sets every one of these on sdk.tracks.uploadTrack and sdk.albums.uploadAlbum today.

Users — fields the indexer only honors on Update

playlist_library, artist_pick_track_id and allow_ai_attribution are read by user_update.go and ignored by user_create.go. That is correct for production — a live client sets them by editing a profile, never at signup. Among the 292,111 users on a production clone never modified after creation, artist_pick_track_id and allow_ai_attribution appear exactly zero times.

A migrated Create is different in kind: it carries an account's final state, not a signup. So migratedUserCreateHandler now accepts them, via three new fields on userState that the production path leaves at their zero values.

// production — unchanged, still ignores them
insertUserWithState(ctx, params, userState{IsAvailable: true})

TestMigratedUserCreate_CarriesUpdateOnlyProfileFields pins both halves: the migration path writes them, the production path does not.

Why not a follow-up User:Update transaction?

That was the first cut. Emitting them on Create instead keeps 637,899 transactions out of genesis and avoids modelling a profile edit that never happened. The indexer change is confined to the migration override, so production create semantics are untouched either way.

Counts

Excluding jsonb columns holding the literal null — these tables use it heavily in place of SQL NULL, and IS NOT NULL counts them:

tracks     producer_copyright_line 1,459   parental_warning_type 1,459
           artists 1,447   rights_controller 1,404   copyright_line 782
           indirect_resource_contributors 782   resource_contributors 686
playlists  artists 143   ddex_release_ids 143
           copyright_line 28   producer_copyright_line 28
users      playlist_library 622,480   artist_pick_track_id 13,935
           allow_ai_attribution 1,484

JSON null values are dropped automatically: unmarshalJSONB decodes them to a nil interface{}, which omitempty omits.

Out of scope

is_original_available (28,827 tracks) is already emitted by the writer and dropped by the indexer, along with seven track DDEX readers this PR now feeds. That is an indexer-side gap that also affects live uploads, and it ships separately.

🤖 Generated with Claude Code