feat: add support for GPU resources

[miguelvr]

Description, Motivation and Context

nvidia.com/gpu is a common resource for many AI applications, but currently, Troubleshoot does not allow users to check the availability of GPUs in their cluster.

I've considered taking a more generic approach with a function resourceName(<resource>), but decided against it because it would either require taking in the units as a parameter resourceName(<resource>, <units>) or default to something like DecimalSI, which could lead to errors for certain resources.

For simplicity, I decided to go with the more explicit approach.

Checklist

• New and existing tests pass locally with introduced changes.
• Tests for the changes have been added (for bug fixes / features)
• The commit message(s) are informative and highlight any breaking changes
• Any documentation required has been added/updated. For changes to https://troubleshoot.sh/ create a PR here

Does this PR introduce a breaking change?

• Yes
• No

[banjoh]

Thanks for your contribution. We'll review the PR

[banjoh]

You took the right direction, but I'm concerned that a GPU device only considers Nvidia hardware. Users who have gpu.intel.com/i915 or gpu.intel.com/xe for example, cannot use this analyser.

[banjoh]

You took the right direction, but I'm concerned that a GPU device only considers Nvidia hardware. Users who have gpu.intel.com/i915 or gpu.intel.com/xe for example, cannot use this analyser.

I don't yet have a clean suggestion that works, but I'd explore combining filters and outcomes.

[xavpaice]

Internal tracking: https://app.shortcut.com/replicated/story/106618/in-cluster-collector-gpu-inventory

[xavpaice]

You took the right direction, but I'm concerned that a GPU device only considers Nvidia hardware. Users who have gpu.intel.com/i915 or gpu.intel.com/xe for example, cannot use this analyser.

I think at this stage we can document the limitation and work with just Nvidia, expanding the scope in a future PR if there's demand.

[miguelvr]

The more generic approach would be fetching by resource name. However it would have to be a function resourceCapacity(nvidia.com/gpu, DecimalSI)

This would cover any custom resource added by any device plugin.

I started on this, but I was having a hard time getting the regex to work as I wanted it to

[xavpaice]

#1162 is related

[chris-sanders]

@miguelvr we were reviewing this and issue #1162 from above and wanted to get your input on this approach.

https://docs.google.com/document/d/1thoEmuHpzf961cHHRreNjVB1rC_u8CVZYBCfxYNg8mY/edit?tab=t.0#heading=h.1q5hq5a3s115

I think that would be a bit more flexible to address the other open issue but want to confirm it would resolve your issue as well. I think it's what you're doing, but just a bit more flexible for the future. I'd like you to review that though and let us know what you think :)