[PyTorch Edge] Remove Original Weight Tensor from QNNPack Sparse Quantized Linear Packed Params (#80473)
We plan to add serialization/deserialization wihout the original weight tensor, so we no longer need to store it
Differential Revision: [D34617321](https://our.internmc.facebook.com/intern/diff/D34617321/)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/80473
Approved by: https://github.com/qihqi
