feat(cluster): expose every tracked replication stream key's position

The remaining reseed defect cannot be diagnosed from the current status surface.
A stream key is a per-LEADER-REGION id (`shard_of_region`), not a shard group, so
a group accumulates one key per leadership it has followed — but every status field
reports only the CURRENT leader's key. A position retained from a previous
leadership is therefore invisible, while the receiver's gap check
(`receiver.rs`: `request_catchup(key, applied + 1)`) will chase ANY key that
received data this round.

That is the blind spot: live tidaldb-0 pulls `from_seqno=13540653`, so some key
sits at 13540652, while the group it reports on converged at 13540661 — and
nothing in the status can say which key that is.

Adds `ReplicationState::applied_by_key` and surfaces it as `applied_by_key` on the
status response. Instrument only: no behavioural change. Same instrument-first move
that turned the previous two defects into one-run diagnoses instead of speculation.

tidal-server/src/cluster/node.rs

@@ -3857,6 +3857,16 @@ impl ShardReplica {
             election_frontier: election_pos.frontier,
             leader_acked: self.leader_acked_frontier(),
             leader_seqno,
+            // Every tracked stream key, not just the current leader's. See
+            // `ReplicationState::applied_by_key` — a key retained from a PREVIOUS
+            // leadership is what the receiver's gap check can keep chasing, and it
+            // was invisible in every other field.
+            applied_by_key: db
+                .replication_state()
+                .applied_by_key()
+                .into_iter()
+                .map(|(k, v)| (u32::from(k.0), v))
+                .collect(),
             reseed_required,
             reseeding,
             self_restart_refused: self.self_restart_refused.load(Ordering::Acquire),
@@ -5223,6 +5233,19 @@ pub struct LocalStatusResponse {
     /// `leader_seqno: 0` means NO INFORMATION, not converged.
     #[serde(default)]
     leader_seqno: u64,
+    /// Every tracked replication stream key with its applied frontier, as
+    /// `(key, applied)` pairs ascending by key.
+    ///
+    /// A key is a per-LEADER-REGION stream id (`shard_of_region`), not a shard
+    /// group, so a group accumulates one key per leadership it has followed. Every
+    /// other field reports only the CURRENT leader's key, which made a position
+    /// retained from a previous leadership unobservable — while the receiver's gap
+    /// check (`receiver.rs`: `request_catchup(key, applied + 1)`) will chase ANY
+    /// key that received data. In the 2026-08-20 incident a node pulled
+    /// `from_seqno=13540653` (a key at 13540652) while the group it reported on sat
+    /// at 13540661, and nothing in the status could say which key that was.
+    #[serde(default)]
+    applied_by_key: Vec<(u32, u64)>,
     /// Whether this node is quarantined with a divergent suffix (m11p4):
     /// fenced from the data plane until reseeded.
     quarantined: bool,

tidal/src/replication/state.rs

@@ -136,6 +136,29 @@ impl ReplicationState {
             .map(|a| a.load(Ordering::Acquire))
     }
 
+    /// Every tracked stream key with its contiguous applied frontier, ascending by
+    /// key.
+    ///
+    /// A key is a per-LEADER-REGION stream id (`shard_of_region`), NOT a shard
+    /// group, so one group accumulates a key per leadership it has followed. Only
+    /// the CURRENT leader's key appears in the per-group status row, which left a
+    /// position tracked under a PREVIOUS leadership completely unobservable — and
+    /// the receiver's gap check issues `request_catchup(key, applied + 1)` for any
+    /// key that received data, current or not. During the 2026-08-20 incident a
+    /// node kept pulling `from_seqno=13540653` (some key at 13540652) while the
+    /// group it reported on sat at 13540661, and there was no way to see which key
+    /// was responsible. This is that instrument.
+    #[must_use]
+    pub fn applied_by_key(&self) -> Vec<(ShardId, u64)> {
+        let mut out: Vec<(ShardId, u64)> = self
+            .applied
+            .iter()
+            .map(|(k, v)| (*k, v.load(Ordering::Acquire)))
+            .collect();
+        out.sort_unstable_by_key(|(k, _)| k.0);
+        out
+    }
+
     /// Whether the batch range `[first, last]` has already been applied for
     /// `shard_id` — its whole span is at or below the contiguous frontier, or it
     /// is recorded verbatim in the `ahead` buffer as an out-of-order apply. The