The Barcode Rank plot shows very high background in your sample.
The cell estimation from
Cell Ranger will look for a sudden drop in the number of UMIs and use this to call cells. In your case, the drop is too much to the right because the empty droplet plateau is almost completely merged with the real cells. If you have already looked at CellBender your case best corresponds to the rightmost example image.
With high background of free mRNA, even empty droplets will contain a significant number of mRNA molecules and become hard to distinguish from real cells.
This likely is a consequence of the single nulcei preparation. To isolate the nuclei, the cells necessarily need to be broken and all the cytoplasmic mRNA is freed contributing to the background. You could try removing this background in future experiments by increasing the number of washing steps after cell lysis.
Using software such as CellBender, SoupX or DropletUtils may help with identifying and reducing the background. CellBender also will rerun the cell estimation step, automatically adjusting the number of called cells.
In any case you can manually reduce the number of cells by filtering based on the minimum number of UMIs or genes. However, the higher the background the more arbitrary such a thresholding strategy will become.