nx_restart_strategy.nx

nx_restart_strategy.nx -- S-CLASS supervision restart strategy (operator: "get those s class exceed stable ... we arent s class stable compared to aws"). PURE decision logic (no syscalls -> unit-gateable offline), grounded in the researched SOTA (knowledge/library/rel_erlang_otp.txt = OTP supervisor restart-INTENSITY; rel_backoff.txt = exponential backoff + jitter; rel_circuit_breaker.txt = contain a failing component). THE GAP IT CLOSES: today nx_hostctl restarts a crashing daemon FOREVER (the gallery/vroom crash-loops spin + spam logs + burn CPU). The S-class rule (Erlang OTP): a child may restart at most MaxR times within MaxT seconds; exceed that and the supervisor STOPS restarting it = QUARANTINE + escalate, so the failure is CONTAINED (bounded blast radius) and a human/alert is engaged instead of an infinite loop. Restarts are spaced by EXPONENTIAL BACKOFF (capped + jittered) so a flapping dependency gets time to recover and N daemons don't thunder-herd. license_tier: ORIGINAL