add inverse_block_masking

[Manishram-ai]

Added inverse_block_masking to masking.py

I tried to keep it like pytorch code using nn.Module. This code now looks similar to other masking implementation.

Thanks!

Edit:

• Improved inverse masking that handles wrap-around, different aspect ratios, and makes sure there are enough visible patches in the block mask
• Also added import statement for math, random and einops
• Class InverseBlockMasking nn.Module is a wrapper to match other masking modules