[Packaging][Release] Reduce disk requirements for linux packaging jobs

[raulcd]

Describe the enhancement requested

Currently we have to perform some disk cleanup on the default GH runners system in order to build Linux packaging jobs, otherwise we get out of space.
Example of occurrences in the past:

• GH-34309: [C++] Disable LTO for aws_lc and s2n-tls #34349 (comment)
• [CI][Packaging] ubuntu-jammy-amd64 packaging job fails due to no space left on device #35066
• [CI][Packaging] ubuntu-jammy packaging job fails due to out of space #35946

On the last issue we temporary removed several things from /usr/local/bin, /usr/local/share and /opt/hostedtoolcache.

There are some ideas for improvements proposed on this comment #35947 (comment)
From the comment

Short term workarounds:

1. Use self-hosted runner like arm64
2. Disable some large components such as S3 and GCP

Long term solution:

1. Build deb packages for large dependencies such as aws-sdk-cpp and google-cloud-cpp separately and use them to build our deb packages

FYI: https://github.com/datafusion-contrib/datafusion-c/ provides deb packages at https://apache.jfrog.io/ui/native/arrow/ubuntu/pool/jammy/main/d/datafusion-c/ . Apache Arrow C++ dependencies can use the same approach.

One more idea: #35961

Component(s)

Packaging, Release

[lriggs]

Are there any updates on work towards a long term solution? I'm running into this problem trying to build arrow jars using the default ubuntu github runner. The problem started recently and I imagine would start to affect other users as well.

Are there any other temporary workaround I could try in the meantime?

[raulcd]

There is not much update yet, as a temporary workaround we do clean some unused binaries and cached packages on the GitHub runner by using the following script: https://github.com/apache/arrow/blob/main/ci/scripts/util_free_space.sh