Add CUDA 11.3 support for manywheel scripts

[janeyx99]

^

[seemethere]

Looks like we're currently blocked on this error:

+ DOCKER_BUILDKIT=1
+ docker build -t docker.io/pytorch/manylinux-builder:cuda11.3 --build-arg BASE_CUDA_VERSION=11.3 --build-arg GPU_IMAGE=nvidia/cuda:11.3-devel-centos7 --target cuda_final -f /raid/eliuriegas/builder/manywheel/Dockerfile /raid/eliuriegas/builder
[+] Building 1.2s (7/7) FINISHED                                                                                                                          
 => [internal] load build definition from Dockerfile                                                                                                 0.0s
 => => transferring dockerfile: 44B                                                                                                                  0.0s
 => [internal] load .dockerignore                                                                                                                    0.0s
 => => transferring context: 2B                                                                                                                      0.0s
 => resolve image config for docker.io/docker/dockerfile:experimental                                                                                0.7s
 => CACHED docker-image://docker.io/docker/dockerfile:experimental@sha256:600e5c62eedff338b3f7a0850beb7c05866e0ef27b2d2e8c02aa468e78496ff5           0.0s
 => [internal] load build definition from Dockerfile                                                                                                 0.0s
 => => transferring dockerfile: 44B                                                                                                                  0.0s
 => [internal] load metadata for docker.io/library/centos:7                                                                                          0.3s
 => ERROR [internal] load metadata for docker.io/nvidia/cuda:11.3-devel-centos7                                                                      0.3s
------
 > [internal] load metadata for docker.io/nvidia/cuda:11.3-devel-centos7:
------
failed to solve with frontend dockerfile.v0: failed to solve with frontend gateway.v0: rpc error: code = Unknown desc = docker.io/nvidia/cuda:11.3-devel-centos7 not found

Maybe we should just maintain these images ourselves, I don't think they'd be too hard to recreate

Actually I think we can use a previous version of the CUDA image since all we actually use the base image for is so that docker run --gpus all <IMAGE> will work correctly

[janeyx99]

Oh okay yea--I also just looked and NVIDIA started adding minor versions to their names, so nvidia/cuda:11.3.0-devel-centos7 exists.