fix: use self.nodes() in Diagram.__init__ to populate nodes_to_show

The custom __iter__ only yields from nodes_to_show, creating a
chicken-and-egg problem: when __init__ used 'for node in self:'
to populate nodes_to_show for a schema source, nodes_to_show was
still empty so __iter__ yielded nothing, leaving nodes_to_show empty.

Fix: replace 'for node in self:' with 'for node in self.nodes()'
which calls the inherited nx.DiGraph node iterator directly,
independent of nodes_to_show.

Fixes test_erd and all Diagram tests that failed due to empty graphs.

Co-Authored-By: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com>

Author: dimitri-yatsenko

src/datajoint/diagram.py

@@ -140,7 +140,7 @@ def __init__(self, source, context=None) -> None:
                     database = source.schema.database
                 except AttributeError:
                     raise DataJointError("Cannot plot Diagram for %s" % repr(source))
-            for node in self:
+            for node in self.nodes():
                 # Handle both MySQL backticks and PostgreSQL double quotes
                 if node.startswith("`%s`" % database) or node.startswith('"%s"' % database):
                     self.nodes_to_show.add(node)