fix(abcip): ALMA review round 1 + ALMD polls InAlarm (not InFaulted)

M1 drop-out is a no-row Unspecified+IsRestatement update (projector case); M2 ack judged over every declared
limit; L1 ActiveAcknowledged for an already-acked new limit; L2 Bad Acked/severity carried/skipped; L3 host
end-to-end test; L4 BOOL member-by-member tests + mixed ALMD/ALMA; L5 detection requires no plain Acked.
ALMD active bit is InAlarm per Rockwell's ALMD reference; InFaulted is input health.

Claude-Session: https://claude.ai/code/session_01Ef2ozxS7QPanGT8uoATbTe

docs/AlarmTracking.md

@@ -125,7 +125,7 @@ Two companions keep the host and driver in step with the carried node:
   deploy, rebuild or not.
 - MTConnect's "last told" record is pruned to the conditions that **left** the subscribed set rather than
   cleared on any set change; re-raising a surviving (carried) condition would un-acknowledge it. AbCip's ALMD
-  last-seen `InFaulted` record is likewise driver-owned and pruned to the set, so a re-subscribe (every
+  last-seen `InAlarm` record is likewise driver-owned and pruned to the set, so a re-subscribe (every
   reconnect / set change) does not re-raise a still-faulted alarm.
 - A routed **client** acknowledgement marks the projector's prior state acknowledged
   (`NativeAlarmProjector.MarkAcknowledged`), so a later driver Clear does not project it unacknowledged.