fix(lists): kind 30005 writes prose into .content, which NIP-51 reserves for private items #7588

Problem

CuratedListService publishing a public curated list writes the description into .content (mobile/lib/services/curated_list_service.dart:1191-1193):

if (list.isPublic) {
  content = list.description ?? 'Curated video list: ${list.name}';
  tags = CuratedListConverter.toEventTags(list);
}

NIP-51 puts a set's description in a tag — "sets can optionally have a title, an image and a description tags" — and reserves .content for the NIP-44-encrypted private-item array. The converter already emits that tag (mobile/packages/curated_list_repository/lib/src/curated_list_converter.dart:190-191), so the prose in .content is both redundant and protocol-wrong.

This is the same defect PR #6966 fixed for kind 10003, where .content had been the literal "Divine global bookmarks".

Data-loss path

NIP-51 allows one list to carry public and private items simultaneously — the spec's own mute-list example does. For a kind-30005 with public e/a tags and another client's ciphertext in .content, unsealItemTags short-circuits before decrypting (mobile/lib/services/curated_list_relay_gateway.dart:484):

if (event.content.isEmpty || _hasPublicItemTags(event.tags)) {
  return const UnsealedItemTags.notSealed();
}

Divine then treats the list as public, and the next edit overwrites .content with the description — destroying private items it never read. Kind 30005 is addressable/replaceable, so the overwrite is total.

Suggested fix

Publish a public list with content: '' and rely on the existing description tag. Carry any existing .content through verbatim when the private items are not what is being changed, mirroring how BookmarkService handles kind 10003 (mobile/lib/services/bookmark_service.dart:722-728).

Note curated_list_converter.dart:107 currently falls back to event.content as the description when reading a non-private list, so the read path needs to keep tolerating legacy events that still carry prose there.

Not measured

The incidence of mixed public+private kind-30005 lists in the wild has not been measured, so the data-loss path is demonstrated from the code rather than from an observed occurrence.