[FEA] Support RAPIDS Spark plugin on ARM

[sameerz]

Is your feature request related to a problem? Please describe.
A clear and concise description of what the problem is. Ex. I wish the RAPIDS Accelerator for Apache Spark would run on ARM

Describe the solution you'd like
Distribute a plugin jar that is functional on ARM-based architecture (including cudf binaries on ARM)

More details:
We need to distribute the spark-rapids-jni jar with 2 different classifers:

1. cuda11-amd64
2. cuda11-aarch64

And Spark-rapids may have 2 profiles,
one is for amd64 to depend on jni's cuda11-amd64 classifier,
another is for aarch64(arm) to depend on jni's cuda11-aarch64 classifier.

For amd64, we build on centos7.
For Arm, we'll build on ubuntu 18.04.
It would be nice if we can build jni jar with Dockerfile.
https://hub.docker.com/layers/nvidia/cuda/11.5.2-runtime-ubuntu18.04/images/sha256-05d99b36cccb63102dbb1f5d9e74daf07d49a62151175693a6146c6d0244dc10?context=explore

Describe alternatives you've considered

Additional context

[pxLi]

Let's keep this open

We will reapply CICD update after developers refactor modules structures for different cpu archs in plugin and JNI repos.

[abellina]

UCX tracking issue: openucx/ucx#8762

[abellina]

Just to be clear @pxLi:

1. Will our classifier be aarch64 or arm64?
2. Is the docker image in the description still the right one?
3. What is the latest/best method to produce arm artifacts that we could share with the UCX developers here Include arm64 in the UCX and JUCX published artifacts openucx/ucx#8762?

[pxLi]

Just to be clear @pxLi:

1. Will our classifier be aarch64 or arm64?
2. Is the docker image in the description still the right one?
3. What is the latest/best method to produce arm artifacts that we could share with the UCX developers here Include arm64 in the UCX and JUCX published artifacts openucx/ucx#8762?

1. I am not sure. As we decide to defer the arm support, the idea should come from spark team
2. Our last try was to build on the arm-VM on aws as it provides us the minimal available versions of LIBC and LIBCXX libs, we were actually not trying to build in docker for our init arm arch artifact
3. We have the PR to revert cpu_arch based build, but I assume that solution was not good enough https://github.com/NVIDIA/spark-rapids/pull/7135/files

[pxLi]

merged #9888, udpating internal pipelines. I will update this ticket no other issue