Fix low accuracy in distilbert-base-uncased

[amalbasaTT]

https://github.com/tenstorrent/pytorch2.0_ttnn/blob/main/docs/models/distilbert-base-uncased/input_variations.md

[amalbasaTT]

Current accuracy is 55.89%.
Low accuracy casues are:

• ttnn.reshape ([Bug Report] Reshaping tiled uint32 tensor changes stored values tt-metal#19223)
• ttnn.eq (ttnn.eq unary produces incorrect results when input tensor is uint32 and tiled and when scalar is 0 tt-metal#20252)

New accuracy is 99.9%.