Update sub_quadratic_attention.py

[nero-dv]

Fixes modules/sub_quadratic_attention.py:205

TODO: maybe we should use torch.empty_like(query) to allocate storage in-advance, and pass slices to be mutated, instead of torch.cat()ing the returned slices

1. Determine the number of query chunks.
2. Calculate the final shape of the res tensor.
3. Initialize the tensor with the calculated shape and dtype, (same dtype as the input tensors, usually)

Can initialize the tensor as a zero-filled tensor with the correct shape and dtype, then compute the attention scores for each query chunk and fill the corresponding slice of tensor.

[AUTOMATIC1111]

I don't think you did any testing on this. np is not imported, and after that is fixed, any generation is just broken because your implementation returns incorrect sizes at dimension 1.