fix(hal/windows): fall back to PowerShell CIM when wmic is unavailable

[rjckkkkk]

Problem

internal/hal/detect_windows.go detects CPU model, RAM, swap, and CPU load by shelling out to wmic. Recent Windows 11 builds no longer ship wmic, so every probe fails:

WARN wmic cpu detection failed: exec: "wmic": executable file not found in %PATH%
WARN wmic RAM detection failed:  exec: "wmic": ...
"cpu": { "model": "", "cores": 32 }     "ram": { "total_mib": 0 }

Result on a Ryzen AI Max+ 395 box: hardware detection (and the onboarding wizard's hardware step) shows no CPU name / no RAM / no VRAM.

Fix

Keep wmic as the primary probe (zero change on machines that have it); when it errors, fall back to powershell Get-CimInstance. Each CIM command emits the same unquoted header\ndata CSV the existing wmic parsers already consume, so the parsing code is unchanged:

Probe	CIM source
CPU	Win32_Processor → Name / NumberOfCores / NumberOfLogicalProcessors / MaxClockSpeed
RAM	Win32_OperatingSystem → TotalVisibleMemorySize / FreePhysicalMemory
Swap	Win32_PageFileUsage → AllocatedBaseSize
CPU load	Win32_Processor.LoadPercentage

Also trims a possible UTF-8 BOM from PowerShell output before parsing, and trims fields defensively.

Tests

detect_windows_test.go (build-tagged windows): with the wmic command absent from the mock runner (→ ErrNotFound), detectCPU/detectRAM fall back to the CIM commands and populate model/cores/threads/freq and total/available/swap. Existing parseWMIC* CSV tests unchanged.

Verified on real Windows 11 hardware (cross-compiled go test -c, run on the box): all hal tests pass, and aima hal detect now reports the CPU model + 128 GiB RAM where it previously returned empty/zero.

Note: these tests can't run on non-Windows CI (//go:build windows); they were executed on a Windows 11 host.

[rjckkkkk]

Closing as a duplicate of #78, which already replaces wmic with PowerShell CIM for CPU/RAM/pagefile/load AND adds AMD GPU (Win32_VideoController) detection — a strict superset of this PR. I reimplemented the CIM CPU/RAM fix here without spotting #78 first; #78 is the more complete fix and should land instead.