Use attribute name as authoritative step identifier in graph

FlowGraph._create_nodes walks dir(cls) by attribute name and stores
nodes[element] = node, but DAGNode.name was derived from the AST def
name. Those agreed only by convention. When a metaclass renames a
function via __name__ (e.g. FunctionSpec synthesizing a step), AST
parsing still reads the original def name, leaving nodes keyed by the
attribute name while node.name reports the def name. FlowSpec._init_graph
then does getattr(cls, node.name) and fails with AttributeError.

Pass the discovered attribute name explicitly to DAGNode in the AST
path, and have DAGNode prefer that over func_ast.name. Both paths
(AST and the sourceless single-step fallback) now key on the same
identifier. No behavior change for flows where def name and attribute
name coincide, which is every ordinary flow.

Author: talsperre

metaflow/graph.py

@@ -65,7 +65,11 @@ def __init__(
         if func_ast is None and name is None:
             raise ValueError("name is required when func_ast is None")
 
-        self.name = func_ast.name if func_ast is not None else name
+        # Prefer explicit `name` (the attribute name under which the step was
+        # discovered) over the AST-derived `def` name. This matters when a
+        # metaclass renames a function via __name__ — the attribute name is
+        # what getattr(cls, node.name) later looks up in FlowSpec._init_graph.
+        self.name = name if name is not None else func_ast.name
         self.source_file = source_file
         # lineno is the start line of decorators in source_file
         # func_ast.lineno is lines from decorators start to def of function
@@ -397,6 +401,7 @@ def _create_nodes(self, flow):
                         is_start_step=is_start,
                         is_end_step=is_end,
                         node_info=getattr(func, "node_info", None),
+                        name=element,
                     )
                 nodes[element] = node
         return nodes