[Op Benchmark] Add Pointwise Conv2d Op Benchmark (#91918)
@bypass-github-export-checks
Pointwise Conv2d is one of the ops which we want to benchmark using different Vulkan Shaders (```conv2d_pw_2x2``` vs ```conv2d_pw_1x1```) with
The configs are copied from Conv2d with the kernel parameter removed.
I considered just using the same configs but ignoring the provided kernel and hardcoding the kernel to 1 when initializing nn.Conv2d, but then in the op benchmark title, it would say kernel=3 even if though that would not be the case.
Differential Revision: [D42303453](https://our.internmc.facebook.com/intern/diff/D42303453/)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/91918
Approved by: https://github.com/mcr229
