examine and diagnose report no GPU and no memory ceiling on WSL

Is your feature request related to a problem? Please describe.

On WSL, rocm examine reports no GPU at all. The GPU probes are skipped on that platform by design, so the device list stays empty and has_amd_gpu reads false even on a healthy machine.

There is also no memory field anywhere in the examination model, so there is no usable-memory figure to report. On an integrated GPU the memory a workload can actually use is smaller than the total the machine advertises, and users have no way to find out what it is before a model fails to load.

Describe the solution you'd like

Report the GPU that WSL can reach, and the memory a workload can actually use, excluding what the platform holds back.

Additional context

ROCm/rocm-cli#340 made WSL its own diagnosis platform but deliberately left the device and memory facts unpopulated, so this is the follow-on.

Two adjacent reports, neither of which covers the examine path:

• ROCm/rocm-cli#276 is the same blindness in the dashboard collector. It also notes that amd-smi metric --json works on WSL ROCDXG, which may be a usable source here.
• ROCm/rocm-cli#332 is a user hitting the downstream symptom when serving.