MBS-12781: Avoid admin accidentally breaking options trees

[reosarevok]

Fix MBS-12781

When editing attributes, relationship types and relationship attributes, you can change their parent type.
There's currently nothing stopping an admin from accidentally selecting the entity itself from the dropdown, which makes MB so confused it doesn't know where to show the entity - so it doesn't. Fun times.

I'm also blocking linking a type as a child of its own child, which also causes similar breaking loop messes. To do that properly we need a recursive query, but the tables are relatively small and these edits happen fairly rarely, so it's probably not an issue.

[mwiencek]

A recursive query isn't much harder, so might be worth doing to prevent future mistakes? (I didn't even know you could set a type as its own parent at present, but here we are. :))

WITH RECURSIVE hierarchy(parent, children) AS (
  SELECT parent, ARRAY[id] AS children
    FROM link_type
  UNION ALL
  SELECT link_type.parent, hierarchy.children || link_type.id
    FROM link_type
    JOIN hierarchy ON link_type.id = hierarchy.parent
   WHERE link_type.parent != any(hierarchy.children)
)
SELECT 1
  FROM hierarchy
 WHERE parent = 123
   AND 456 = any(children)
 LIMIT 1;

I guess Data::LinkType doesn't consume Data::Role::OptionsTree for some reason, but we could just duplicate the query in Data::LinkType for now.

[reosarevok]

Ok then! I updated the PR to use the recursive query - it was breaking for LinkAttributeType because that has a weird _table, but I just hardcoded that one.

[mwiencek]

It would be nice to remove the link type duplication in the future (the form validation can probably already be a role), but this works for now. :)

[reosarevok]

Updated this to use %ENTITIES. We could move the validation to a role indeed - the only reason not to might be that LinkAttributeType makes extra checks for instruments, but I guess that could be either a second after validate or a check for model inside the role. Which of the two sounds better?

[reosarevok]

@brainzbot, retest this please

[yvanzo]

Didn’t review yet but this can be rebased after the pull request #2777 that you updated today.