_gpu_dxg_r6_hwq.nx

_gpu_dxg_r6_hwq.nx -- SOVEREIGN-GPU R6: make the real RTX 5080 EXECUTE via the HARDWARE-QUEUE path. Now rides the CONSOLIDATED dxg device shim (nx_dxg) instead of inline-copying the control plane. HYPOTHESIS (grounded in the open WSL2-Linux-Kernel dxgkrnl ABI): R5a used LX_DXSUBMITCOMMAND (packet- scheduling). Modern NVIDIA on WSL uses HARDWARE SCHEDULING -> packet-scheduled submit is ACCEPTED but never dispatched. The HWS path = D3DKMTCreateHwQueue + D3DKMTSubmitCommandToHwQueue (in nx_dxg). CreateHwQueue returns a MONITORED PROGRESS FENCE the GPU signals on completion -> a clean dispatch/exec signal. RESULT (recorded): CreateHwQueue -EINVAL = host NVIDIA KMD rejects (needs proprietary priv_drv_data) -> the dxg path is the proprietary-handshake dead-end; the sovereign route is the open-kernel-module/bare-metal path on native-Linux/passthrough (see project-cuda-exceed-roadmap). NISHI-ONLY: we emit the raw pushbuffer; no libcuda/Mesa. RISK: a malformed HW submit may TDR; WSL recovers. license_tier: ORIGINAL