collatedFileOperation: preferentially collect all data in the simulation thread

so the write thread does not have to do any parallel communication.  This avoids
the bugs in the threading support in OpenMPI.

Patch contributed by Mattijs Janssens
Resolves bug-report https://bugs.openfoam.org/view.php?id=2669

src/OpenFOAM/db/IOobjects/decomposedBlockData/decomposedBlockData.H

@@ -169,6 +169,32 @@ public:
             const UPstream::commsTypes commsType
         );
 
+        //- Helper: gather single label. Note: using native Pstream.
+        //  datas sized with num procs but undefined contents on
+        //  slaves
+        static void gather
+        (
+            const label comm,
+            const label data,
+            labelList& datas
+        );
+
+        //- Helper: gather data from (subset of) slaves. Returns
+        //  recvData : received data
+        //  recvOffsets : offset in data. recvOffsets is nProcs+1
+        static void gatherSlaveData
+        (
+            const label comm,
+            const UList<char>& data,
+            const labelUList& recvSizes,
+
+            const label startProc,
+            const label nProcs,
+
+            List<int>& recvOffsets,
+            List<char>& recvData
+        );
+
         //- Write *this. Ostream only valid on master. Returns starts of
         //  processor blocks
         static bool writeBlocks
@@ -177,6 +203,12 @@ public:
             autoPtr<OSstream>& osPtr,
             List<std::streamoff>& start,
             const UList<char>&,
+
+            const labelUList& recvSizes,
+
+            const bool haveSlaveData,       // does master have slaveData
+            const List<char>& slaveData,    // optional slave data (on master)
+
             const UPstream::commsTypes,
             const bool syncReturnState = true
         );