docs(world-model): outstanding work as handoff

[buremba]

Adds a prioritized "Outstanding work" section to `docs/plans/world-model.md` so the next session has a concrete checklist.

Principles for the items below

• Reuse existing surfaces. No new pages, MCP tools, or components.
• Widen, don't replace. Type pickers grow one badge column; read sites widen with the same JOIN pattern feat(world-model): cross-org references — schema search path + write guard #374/feat(world-model): cross-org relationship_types + catalog discovery #377 already use.
• Defer the heavy stuff. Items not needed for current usage (TOCTOU rewrite, slug→id rules migration, claims, federation) stay out of the active list.

Items, ordered by what unblocks usage

1. Schema CRUD cross-org (backend, ~80 LOC) — `manage_entity_schema` list/get widens; rows include origin org. Prerequisite for items 2-3.
2. Frontend type pickers show cross-org with read-only badge (~30 LOC) — purely visual on existing dropdown.
3. Discovery via existing org dropdown — group public catalogs separately. No global search bar (agents already have `search_knowledge`).
4. Read-side cross-org tolerance (~50 LOC) — `manage_entity.get`, `resolve_path`, public pages widen the same way `fetchEntityById` did in feat(world-model): cross-org relationship_types + catalog discovery #377.
5. Visibility-flip safety (~10 LOC) — refuse `public → private` when referenced. Cleaner than retroactive revalidation.
6. Catalog curation — data pass, no PR. Existing public orgs have demo cruft.
7. `is_catalog` flag — deferred until visibility can be mutated by regular users.
8. `entity_relationship_type_rules` slug → id — deferred until first cross-catalog rule conflict.
9. Contribution flow — premature; build when someone has data worth contributing.
10. TOCTOU on type lookup — deferred; semantic-not-corrupting.

Test plan

• `bun run check` clean