Add LoRA Alpha parameter to modules that support enable_lora

[b05505027]

I suggest we can add an alpha parameter in the enable_lora function. Currently, only the rank can be specified in the function.

Typically, LoRA decomposes the weight update into two matrices A and B and then multiply their product by α/rank to control its magnitude.

For backward compatibility, we can set the default alpha = rank. That way, if you were already using LoRA with a given rank, setting the same rank for alpha reproduces the old behavior exactly.

And I think this could be implemented by multiplying the product of matrix A and B by (self.alpha / self.rank) when calling the kernel.

[divyashreepathihalli]

Please feel free to open a PR for this!