Add CUDA build + packaging workflow

[mattkjames7]

Description

Added Dockerfile.cuda specifically for CUDA build and updated packaging workflows to use it.

A manual build seems to show CUDA as available, e.g.:

matt@matt-MS-7B86:/media/raid/Work/github$ocker run --rm --name mage-cuda -p 7687:7687 --gpus all -d memgraph/memgraph-mage:3.5.1-cuda --telemetry-enabled=False --log-level=TRACE
de3bd647563738c8040dcab120068c67cf78c324e9530ce438fd01fcc9f02236
matt@matt-MS-7B86:/media/raid/Work/github$ docker ps
CONTAINER ID   IMAGE                               COMMAND                  CREATED         STATUS         PORTS      NAMES
de3bd6475637   memgraph/memgraph-mage:3.5.1-cuda   "/usr/lib/memgraph/m…"   4 seconds ago   Up 3 seconds   7687/tcp   mage-cuda
0ab1328faca6   moby/buildkit:buildx-stable-1       "buildkitd --allow-i…"   3 months ago    Up 10 hours               buildx_buildkit_stupefied_jones0
matt@matt-MS-7B86:/media/raid/Work/github$ docker exec -it -u memgraph mage-cuda bash
memgraph@de3bd6475637:/$ python3
Python 3.12.3 (main, Aug 14 2025, 17:47:21) [GCC 13.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import torch
>>> torch.cuda.is_available()
True
>>> torch.cuda.get_device_name(0)
'NVIDIA GeForce RTX 4070 Ti SUPER'
>>> 

• CUDA: 12.6
• PyTorch: 2.6.0

TODO:

• Fine-tune the torch-* install to use prebuilt packages for whichever version of torch we end up building with (otherwise building takes forever, as these packages seem to require building from source).
• Select torch version - current version used in MAGE is 2.6.0; current stable version is 2.8.0 - recommendation: use 2.6.0 until dgl is removed.
• Select which version(s) of CUDA we want to support. Current release is 13.0, but the latest supported by torch==2.8.0 is 12.9; recommendation: use version 12.6 as it is supported by all versions of torch from 2.6.0 to 2.8.0, and it still has support for older drivers (>=525 to <580) and GPUs (Pascal, Maxwell).
• ROCm
• Test module to see if it can use the GPU
• Test multiple GPUs
• Revert changes to Dockerfile.release
• Install the https://pypi.org/project/memgraph-toolbox/

Pull request type

• Bugfix
• Algorithm/Module
• Feature
• Code style update (formatting, renaming)
• Refactoring (no functional changes, no api changes)
• Build related changes
• Documentation content changes
• Other (please describe):

Related issues

Delete if this PR doesn't resolve any issues. Link the issue if it does.

######################################

Reviewer checklist (the reviewer checks this part)

Module/Algorithm

• Core algorithm/module implementation
• Query module implementation
• Tests provided (unit / e2e)
• Code documentation
• README short description

Documentation checklist

• Add the documentation label tag
• Add the bug / feature label tag
• Add the milestone for which this feature is intended
  • If not known, set for a later milestone
• Write a release note, including added/changed clauses
  • [Release note text]
• Link the documentation PR here
  • [Documentation PR link]

[sonarqubecloud]

Quality Gate failed

Failed conditions
15 Security Hotspots
E Security Rating on New Code (required ≥ A)

See analysis details on SonarQube Cloud

Catch issues before they fail your Quality Gate with our IDE extension SonarQube for IDE

[gitbuda]

This is working (I've built the image and run it on GPUs). Merging into my branch, I think we should polish the API a bit.