[quantized] Implement 3d convolution in qnnpack (#66350)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/66350
Implements conv3d for QNNPACK by writing another kernel for the indirection buffer in 3 dimensions. Modifies all structs to take depth, with depth = 1 indicating 2d operation. gemm and conv (non transpose) work, next up is depthwise and tranpose.
ghstack-source-id: 140152440
Test Plan: test/quantization
Reviewed By: kimishpatel
Differential Revision: D30858693
fbshipit-source-id: 883cca8ec53b9e15ab4b9473c6cc042e3d049d9c
