Intended behaviour of sub-graphs in a compiled graph

[fazam0616]

When compiling a sub-graph within a larger graph, under specific conditions, the START and END nodes of the sub-graph are included in the larger graph. When the START or END nodes have more than 1 edge outgoing or incoming, respectively, the nodes are included. They are omitted otherwise. Is this intended behaviour? I assume so, as it adds clarity to the fully compiled graph. However I assume there is some kind of performance impact with adding redundant nodes. Would an optional flag to disable the redunadant organizational nodes when compiling be a good idea?
This is inspired by this issue:
/issues/1676

[fazam0616]

The behaviour seems to be unintended: langchain-ai/langchain#28411