Avoid calling the get_typlen multiple times (#256)

This commit also changes the column type length judgment in
ReadChunkGroupNextVector().  According to Postgres code, not all
platforms have sizeof(Datum) == 8 [1], so it might be better to
use sizeof(Datum) instead of the magic number 8.

[1] https://github.com/postgres/postgres/blob/master/src/include/postgres.h#L58

columnar/src/backend/columnar/columnar_reader.c

@@ -2239,10 +2239,10 @@ ReadChunkGroupNextVector(ChunkGroupReadState *chunkGroupReadState, Datum *column
 
 
 				/* 
-				 * For data types which have len less or equal 8 we can
+				 * For data types which have len less or equal sizeof(Datum) we can
 				 * use `store_att_byval` function.
 				 */
-				if (vectorColumn->columnTypeLen <= 8)
+				if (vectorColumn->columnTypeLen <= sizeof(Datum))
 				{
 					store_att_byval(writeColumnRowPosition,
 									chunkGroupData->valueArray[columnIndex][rowIndex],

columnar/src/backend/columnar/vectorization/columnar_vector_types.c

@@ -65,7 +65,7 @@ CreateVectorTupleTableSlot(TupleDesc tupleDesc)
 		int16 columnTypeLen = get_typlen(columnTypeOid);
 
 		int16 vectorColumnTypeLen = 
-			columnTypeLen == -1 ?  sizeof(Datum) : get_typlen(columnTypeOid);
+			columnTypeLen == -1 ?  sizeof(Datum) : columnTypeLen;
 
 		/* 
 		 * We consider that type is passed by val also for cases where we have 
@@ -168,4 +168,4 @@ CleanupVectorSlot(VectorTupleTableSlot *vectorSlot)
 
 	memset(vectorSlot->keep, true, COLUMNAR_VECTOR_COLUMN_SIZE);
 	vectorSlot->dimension = 0;
-}
+}