fix(gh-actions): Make docker and release actions work

[vponomaryov]

Fixes:

• Before docker ARM image building was failing running in Qemu. Now it gets build on a dedicated ARM-based runner. After both builds (amd, arm) are completed we create manifest refering to the built digest/sha256 of both images.
• Before we were trying to create release from 2 artifact builders it was racing and second one was failing the pipeline.
• Also, before it was not expected that release may exist. Redo logic to allow a release be created as well as it's tag.

TODO:

• fix getting of upload URL

[vponomaryov]

Images, AMD and ARM, were created here: https://hub.docker.com/r/scylladb/hydra-loaders/tags?name=latte

Screenshot:

Those images were created exactly by these github actions.
It was created there because I have "write" creds there using my account.
TODO: need to make our QA docker user have "read/write" access to the "scylladb/latte" docker registry/repo.

Github action results which built those images are here: https://github.com/scylladb/latte/actions/runs/13213653580/job/36890378335

As it is written above, left to fix getting of the "upload url" for uploading built binaries.

@fruch , @CodeLieutenant do we really want to compile binaries and attach it to releases?
I am asking just because I am aware about cases where some people use either docker or compile themselves.
Since these binaries depend on system libraries the compilation in place sounds reasonable.

Time notes:

• The docker AMD image build takes about 4m30s
• The docker ARM image build takes about 5m50s
• Binary compilation for ubuntu-22.04 on AMD takes about 3m50s.
• Other binaries compilation takes more.

So, do we want to double the spent time for building binaries which are unknown to be used by anyone?