Y2B-250 — Restore ARM time32 nanosleep for the diagnostic heartbeat

Depends on Y2B-245 #20. Owner hardware evidence: renderer, PID1, proc/sysfs mounts work; first sleep returns -38 (ENOSYS), ending at BEAT: 0 / STOP: SLEEP ERROR.

Linux v6.18 ARM syscall 162 maps to sys_nanosleep_time32, compiled only with CONFIG_COMPAT_32BIT_TIME. The tested allnoconfig output disables it; the linked symbol is a weak alias to sys_ni_syscall. Enable only that config option. CONFIG_POSIX_TIMERS is not the dependency and remains disabled.

Acceptance

• Record the exact owner-reported screen and distinguish it from host tests.
• Resolved config enables time32; linked table slot 162 targets the real function,

  not the ENOSYS fallback. A regression rejects an actual table redirected to ENOSYS.

• One clean locked build; targeted syscall and production PID1 regression tests.
• Existing D08/layout/BOOTIMG checks pass, without relaxing any bound; record hash/size.
• CPU0, static DT, watchdog stop, inherited framebuffer guards, initramfs and

  preloader/LK/partition layout remain unchanged. No flash, M2 or new subsystem.

• Trust previously audited source/toolchain/recovery inputs; no repeated full

  source/provenance/recovery audit or second reproducibility build for this fix.

Expected next trial: advance beyond sleep error to increasing BEAT/FRAME and runtime fields; after 50 successful sleeps, STOP/RESTORE. Hardware behavior is pending a separately authorized test; offline validation does not prove IRQ timing.

Completed offline at dff1e82; result, hash and targeted validation. All acceptance checks passed. Next device trial needs separate authorization.