In the U.S., place=* tags on boundary relations aren’t a best practice in the first place, since it conflates legal classification with a human geography notion of a place’s importance. The legal status is a better fit for a boundary’s border_type=*, while the label node should have place=* indicating something potentially different than the legal classification. I think the general sentiment for years was basically not to worry about the accuracy of place=* tags on boundaries, as they came from an old import that had plenty of other higher-priority things to fix. But it probably would’ve been more responsible to delete the tag rather than allow it to mislead data consumers.
Pennsylvania also has boroughs, but they mean something different than in New York: like towns but with a different form of government. The TIGER import tagged their boundary relations with border_type=borough place=borough, while the separate GNIS import tagged their label nodes as place=village. If we keep tagging the borough boundaries with place=*, then it should be consistent with the label node.
Unlike border_type=*, place=* is expected to have a roughly similar meaning across jurisdictions, so I agree with retagging the New York boroughs’ label nodes to place=borough, though place=suburb doesn’t sound that far off either. Ideally, we’d remove the place=* tags from boundaries altogether, but this requires ensuring that each place=* point is actually a label member of any boundary relation that corresponds to it. Unfortunately, there seem to be plenty of boundary relations for municipalities that lack label members.