Fix low accuracy in swin_v2_b, swin_v2_s and swin_v2_t (38.3, 63.36, 59.87)

[amalbasaTT]

https://github.com/tenstorrent/pytorch2.0_ttnn/tree/main/docs/models/swin_v2_b

[nemanjagrujic]

@ayerofieiev-tt Here the problem is in broadcast for elementary binary ops: sub, mul, div. Covered by this:

ticket:tenstorrent/tt-metal#18546

When ttnn.mul, ttnn.sub, ttnn.div are exchanged with:

ttnn.experimental.sub
ttnn.experimental.mul
ttnn.experimental.div 

This networks work fine.