Evaluate using systeminformation for GPU detection (variant selection)

## Context
PR #390 added systeminformation for telemetry enrichment but intentionally kept the original PCI vendor ID approach (WMI, lspci, sysfs, nvidia-smi) for GPU detection / variant selection.

The gpus array in telemetry now reports all GPU controllers from si.graphics() alongside the detected gpu_vendor from the legacy path. This data can be used to compare what si would pick vs what the current detection picks.

## Task
1. Collect telemetry data comparing gpu_vendor (legacy detection) against the gpus array (si data) across real user systems
2. Identify any cases where si would produce a different/wrong result (e.g., picking Intel iGPU over NVIDIA dGPU on laptops, Intel Mac edge cases)
3. If si proves reliable, consider migrating detectGPU() to use si.graphics() as the primary detection path

## Key concerns to validate
- Multi-GPU systems (laptop with iGPU + dGPU)
- macOS Intel Macs (should return null, not amd/intel)
- WSL environments (si may not have PCI access)
- Systems where si vendor strings don't match expected patterns

Follow-up to #380.