perf: Move mutex to the partition level (#5486)

This change introduces two key optimizations:
* Mutex scope reduction: Limits the lock to individual partitions within `TaggedCache`, reducing contention.
* Decoupling: Removes the tight coupling between `LedgerHistory` and `TaggedCache`, improving modularity and testability.

Lock contention analysis based on eBPF showed significant improvements as a result of this change.

include/xrpl/basics/partitioned_unordered_map.h

@@ -277,6 +277,12 @@ public:
         return map_;
     }
 
+    partition_map_type const&
+    map() const
+    {
+        return map_;
+    }
+
     iterator
     begin()
     {
@@ -321,6 +327,12 @@ public:
         return cend();
     }
 
+    std::size_t
+    partition_index(key_type const& key) const
+    {
+        return partitioner(key);
+    }
+
 private:
     template <class T>
     void