[ML][Data Frame] have sum map to a double to prevent overflows (#43213)

Author: benwtrent

x-pack/plugin/data-frame/src/main/java/org/elasticsearch/xpack/dataframe/transforms/pivot/Aggregations.java

@@ -34,7 +34,7 @@ enum AggregationType {
         VALUE_COUNT("value_count", "long"),
         MAX("max", SOURCE),
         MIN("min", SOURCE),
-        SUM("sum", SOURCE),
+        SUM("sum", "double"),
         GEO_CENTROID("geo_centroid", "geo_point"),
         SCRIPTED_METRIC("scripted_metric", DYNAMIC),
         WEIGHTED_AVG("weighted_avg", DYNAMIC),

x-pack/plugin/data-frame/src/test/java/org/elasticsearch/xpack/dataframe/transforms/pivot/AggregationsTests.java

@@ -34,9 +34,9 @@ public void testResolveTargetMapping() {
         assertEquals("half_float", Aggregations.resolveTargetMapping("min", "half_float"));
 
         // sum
-        assertEquals("int", Aggregations.resolveTargetMapping("sum", "int"));
         assertEquals("double", Aggregations.resolveTargetMapping("sum", "double"));
-        assertEquals("half_float", Aggregations.resolveTargetMapping("sum", "half_float"));
+        assertEquals("double", Aggregations.resolveTargetMapping("sum", "half_float"));
+        assertEquals("double", Aggregations.resolveTargetMapping("sum", null));
 
         // geo_centroid
         assertEquals("geo_point", Aggregations.resolveTargetMapping("geo_centroid", "geo_point"));