feat(win): better gpu detection

[axel7083]

What does this PR do?

Remove the old GPUManager code detecting the Nvidia GPU using nvidia-smi and replace it with a generic worker system.

This PR is adding the WinGPUDetector worker. This worker is responsible of detecting the GPUs available on windows.

How

We need two information, the name and the VRAM available for each GPU, sadly windows is crap¹²³, and they don't fix their code¹, so we have to read the registry⁴ to access the exact value of the VRAM for each GPU detected.

Reading the registry could feel scary but we are simply following the official Microsoft documentation here, with long term support values. Therefore not too scary :)

Other solution do not provide accurate value³.

Screenshot / video of UI

What issues does this PR fix or reference?

Related to https://github.com/containers/podman-desktop-internal/issues/270

Fixes #1144

How to test this PR?

• Unit tests has been added

Manually

Adding the following in the studio.ts file (activate function)

const gpuManager = new GPUManager(this.#panel?.webview);
gpuManager.collectGPUs().then((gpus) => {
  console.log('Collected GPUs', gpus);
}).catch((err: unknown) => {
  console.error('Something went wrong while tryng to collect GPUs', err);
});

Example output

[ai-lab] Collected GPUs [ { model: 'NVIDIA GeForce GTX 1060 6GB', vram: 6442450944 } ]

Footnotes

1. Msinfo32.exe reports an unexpected value for the Display Adapter RAM when the graphics adapter has 2 GB or more of dedicated video memory https://learn.microsoft.com/en-us/troubleshoot/windows-client/setup-upgrade-and-drivers/msinfo32-report-wrong-display-adapter-ram ↩ ↩²

2. Windows is crap https://superuser.com/questions/1461858/fetch-correct-vram-for-gpu-via-command-line-on-windows ↩

3. Windows is crap https://stackoverflow.com/questions/68274009/wmi-win32-videocontroller-ram-4gb-limit ↩ ↩²

4. Windows registry values https://learn.microsoft.com/fr-fr/windows-hardware/drivers/install/system-defined-device-setup-classes-available-to-vendors ↩

[lstocchi]

Tested and my output is model: "Intel(R) UHD Graphics\u0000" vram: undefined which is in line with what i have since my gpu does not have dedicated ram.

LGTM