[PyTorch Edge] Fix ao::sparse::BCSR missing in qlinear serialize and deserialize when USE_FBGEMM and USE_PYTORCH_QNNPACK are not set (#81256)
Summary:
ao::sparse::BCSR is defined in packed_params.h, which is not included when USE_FBGEMM and USE_PYTORCH_QNNPACK are both not set. Fix this problem by surrounding the functions using ao::sparse::BCSR in qlinear_serialize and qlinear_deserialize with ```#ifdef USE_FBGEMM```.
For more context, see:
- https://github.com/pytorch/pytorch/pull/81081
- https://github.com/pytorch/pytorch/issues/81178
Test Plan:
Build successful
```time USE_CUDA=0 BUILD_CAFFE2_OPS=0 USE_XNNPACK=0 USE_FBGEMM=0 USE_DISTRIBUTED=0 USE_MKLDNN=0 USE_QNNPACK=0 BUILD_TEST=0 USE_GOLD_LINKER=1 USE_OPENMP=0 USE_PYTORCH_QNNPACK=0 DEBUG=1 python setup.py develop```
Differential Revision: D37757170
Pull Request resolved: https://github.com/pytorch/pytorch/pull/81256
Approved by: https://github.com/kimishpatel, https://github.com/ezyang
