2026-08-31 (HUD fields) - GPU field added, DISK wired via statfs, TEMP left honestly unknown

From two HUD questions on a hardware screenshot. (1) GEN showed ps4_mode and dropped the graphics-driver evidence: on a ps4_game the current-gen AGC driver is refused, so the generation verdict collapses to unknown (D255) - but gnm plainly resolves (it renders the report; 165-gnm passes). Added a separate GPU field (obs_gpu_drivers in generation.c, reusing the existing gnm/agc markers) reporting the driver - gnm here - as a fact independent of the console-generation question. (2) DISK wired through statfs: free = f_bavail(0x30) × f_bsize(0x10), read by offset from an over-sized buffer (FreeBSD sys/mount.h, stable offsets; D008 discipline). statfs moved from the mined corpus to platform.h (a const-char census entry and a callable symbol collide) - corpus regenerated (libkernel 789→788), counts/BACKLOG regenerated. TEMP left unknown: no temperature query exists in the corpus, and inventing a thermal signature is exactly what D008 forbids. NET/KEY left as-is (serve-mode fields, correctly absent on a suite run). New sysinfo key gpu added to OUTPUT.md. All affected files compile clean under the strict set; make module builds (mkmodule places statfs); census/counts/decisions/doccheck/guards green. D272. DISK's path (/download0) and offsets want a hardware run to confirm.