perf: Remove SHAMap overhead on hot paths

- SHAMapNodeID derived from CountedObject, so every construction,
  copy and destruction updated one process-wide atomic counter. Tree
  walks create one per level, so all threads contended on that counter.
  The count was only used for get_counts diagnostics.
- NodePathStack used std::stack's default std::deque, which allocates
  on first push. Use a small_vector with 16 inline entries instead.
- fetchNodeNT canonicalized a node that fetchNodeFromDB had already
  canonicalized, which took the TreeNodeCache lock a second time.

include/xrpl/shamap/SHAMap.h

@@ -17,6 +17,8 @@
 #include <xrpl/shamap/SHAMapNodeID.h>
 #include <xrpl/shamap/SHAMapTreeNode.h>
 
+#include <boost/container/small_vector.hpp>
+
 #include <condition_variable>
 #include <cstddef>
 #include <cstdint>
@@ -518,7 +520,9 @@ private:
         }
 
     private:
-        std::stack<std::pair<SHAMapTreeNodePtr, SHAMapNodeID>> stack_;
+        // Paths are rarely deeper than 16 nodes, so the inline buffer avoids the heap.
+        using Entry = std::pair<SHAMapTreeNodePtr, SHAMapNodeID>;
+        std::stack<Entry, boost::container::small_vector<Entry, 16>> stack_;
     };
 
     using DeltaRef =

include/xrpl/shamap/SHAMapNodeID.h

@@ -1,6 +1,5 @@
 #pragma once
 
-#include <xrpl/basics/CountedObject.h>
 #include <xrpl/basics/base_uint.h>
 
 #include <compare>
@@ -16,7 +15,7 @@ namespace xrpl {
 /**
  * Identifies a node inside a SHAMap
  */
-class SHAMapNodeID : public CountedObject<SHAMapNodeID>
+class SHAMapNodeID
 {
 private:
     UInt256 id_;
@@ -98,8 +97,6 @@ public:
      *
      * Only a *defaulted* operator<=> implicitly declares a defaulted
      * operator==; the one above is user-provided, so == has to be written.
-     * It cannot be defaulted either, because a defaulted == would also compare
-     * the CountedObject base, which is not equality comparable.
      */
     bool
     operator==(SHAMapNodeID const& n) const

src/libxrpl/shamap/SHAMap.cpp

@@ -244,12 +244,10 @@ SHAMap::fetchNodeNT(SHAMapHash const& hash, SHAMapSyncFilter const* filter) cons
 
     if (backed_)
     {
+        // fetchNodeFromDB has already canonicalized the node.
         node = fetchNodeFromDB(hash);
         if (node)
-        {
-            canonicalize(hash, node);
             return node;
-        }
     }
 
     if (filter != nullptr)