Speed up stored field spec management

When you load stored fields from lucene you have get all of your ducks
in a row first or it'll be really slow. You need to get a list of all of
the fields you want so you can ignore the ones that you don't. We do
this with the `StoredFieldsSpec` which is immutable and has a `merge`
method. It's quite common to merge a few dozen of these specs together
to prepare for the fetch phase or for a new segment when loading fields
from ESQL.

When I was loading thousands of fields in ESQL I noticed that the merge
was slowing things down marginally. This skips a big chunk of the merge
in the common case that we don't have to load any named stored fields.

Author: nik9000

server/src/main/java/org/elasticsearch/search/fetch/StoredFieldsSpec.java

@@ -42,8 +42,13 @@ public StoredFieldsSpec merge(StoredFieldsSpec other) {
         if (this == other) {
             return this;
         }
-        Set<String> mergedFields = new HashSet<>(this.requiredStoredFields);
-        mergedFields.addAll(other.requiredStoredFields);
+        Set<String> mergedFields;
+        if (other.requiredStoredFields.isEmpty()) {
+            mergedFields = this.requiredStoredFields;
+        } else {
+            mergedFields = new HashSet<>(this.requiredStoredFields);
+            mergedFields.addAll(other.requiredStoredFields);
+        }
         return new StoredFieldsSpec(
             this.requiresSource || other.requiresSource,
             this.requiresMetadata || other.requiresMetadata,