[3.14] gh-150868: Fix sum() deallocating a float subclass as an exact float (GH-151063)

[3.14] gh-150868: Fix sum() deallocating a float subclass as an exact float (GH-151063) #217546