Update OFT to fix merge bugs

[Zeju1997]

changes for BOFT

• fixing small errors in comments of BOFT
• making Conv2D operation consistent with Linear

changes for OFT

• [Important!] fixing the incorrect merging of OFT, it is not an implementation error, but more of an incorrect understanding of how orthogonal fine-tuning works, resulting in a wrong way of merging adapters
• update config.py/layer.py/model.py file to be consistent with other peft methods
• adding additional paramters oft_block_size, oft_dropout, bias
• fixing module dropout implementation error (in orthogonal finetuning, dropout is not the same as LoRA)
• fixing errors in merge and unmerge function in OFT
• making Conv2D operation consistent with Linear