fix: parse vulkan VRAM from text

[eglia]

Description
This PR fixes the vulkan VRAM parsing. As far as I can tell, the current implementation is nonsense, because vulkaninfo --json does not write to stdout:

[-j, --json]         Produce a json version of vulkaninfo output conforming to the Vulkan
                     Profiles schema, saved as 
                     "VP_VULKANINFO_[DEVICE_NAME]_[DRIVER_VERSION].json"
                     of the first gpu in the system.

Also, on my machine the json info did not actually contain the heap information, which is present in the text output.
This PR therefore changes to vulkaninfo --text and parses the output. On my 32GB AMD GPU, this correctly detects the available VRAM.
New output:
Total available VRAM vram=34208743424 caller={caller.file="/build/pkg/system/state.go" caller.L=79 }
Old output:
Total available VRAM vram=0 caller={caller.file="/build/pkg/system/state.go" caller.L=79 }

This PR fixes #9289

Notes for Reviewers

Signed commits

• Yes, I signed my commits.

[Copilot]

Pull request overview

Note

Copilot was unable to run its full agentic suite in this review.

This PR fixes Vulkan VRAM detection by switching from vulkaninfo --json to parsing vulkaninfo --text, since the JSON output is not emitted to stdout and may omit heap info.

Changes:

• Call vulkaninfo --text instead of --json for GPU memory fallback detection
• Add a text parser that extracts device name/type and sums device-local memory heaps
• Remove the prior JSON schema-based parsing logic

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[eglia]

This is not true. This is the output of vulkaninfo --text | grep deviceType:

        deviceType        = PHYSICAL_DEVICE_TYPE_DISCRETE_GPU
        deviceType        = PHYSICAL_DEVICE_TYPE_CPU