[quantized] Replace conv_p with convolution_op in qnnpack (#65783)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/65783
convolution_op makes conv_param struct redundant, since it contains all the params of conv_param and more. We don't need to pass both structs to qnnpack or hold both in the packed weights, let's just hold convolution_op.
This makes it easier to implement 3dconv since we won't have to template two structs. The conv_param struct is left in existence since tests rely on it to set up the convolution.
ghstack-source-id: 139479651
(Note: this ignores all push blocking failures!)
Test Plan: ci
Reviewed By: kimishpatel
Differential Revision: D30738727
fbshipit-source-id: e6d39644357b99d3b7491ae8a7066bf107eb8b9e
