Optimize start position of DOMPage findRecord()

At least in the use case that we have, we tend to scan the page for the same record, or one that comes later than it in position. So instead of beginning each scan from 0, record the last position which succeeded, and start from there. Loop round to scan the unscanned region if that fails.

That seems to save us a significant amount of time;
DOMFile$DOMPage$findRecord was 25% of CPU in our “before” workload, it is 3% after.
DOMFile.findRecord was 41% of CPU, and is now 23% of CPU.
Query timings have gone from about 4.9s to about 3.6s

exist-core/src/main/java/org/exist/storage/dom/DOMFile.java

@@ -29,6 +29,7 @@
 import java.text.NumberFormat;
 import java.util.ArrayList;
 import java.util.List;
+import java.util.concurrent.atomic.AtomicInteger;
 import javax.xml.stream.XMLStreamConstants;
 import javax.xml.stream.XMLStreamException;
 import javax.xml.stream.XMLStreamReader;
@@ -3048,6 +3049,8 @@ protected final class DOMPage implements Cacheable {
         // set to true if the page has been removed from the cache
         boolean invalidated = false;
 
+        AtomicInteger lastFound = new AtomicInteger(0);
+
         DOMPage() {
             this.page = createNewPage();
             this.pageHeader = (DOMFilePageHeader) page.getPageHeader();
@@ -3093,10 +3096,33 @@ Page createNewPage() {
             }
         }
 
+        /**
+         * Optimize scanning for records in a page
+         * Based on the fact that we are looking for the same thing again,
+         * Or we are looking for something after the last thing.
+         *
+         * So, start the scan where we left off before.
+         *
+         * @param targetId the tuple id we are looking for in the page
+         *
+         * @return a record describing the tuple, if we found it, otherwise null
+         */
         RecordPos findRecord(final short targetId) {
-            final int dlen = pageHeader.getDataLength();
+
+            int startScan = lastFound.get();
+            RecordPos rec = findRecordInRange(targetId, startScan, pageHeader.getDataLength());
+            if (rec == null) {
+                rec = findRecordInRange(targetId, 0, startScan);
+            }
+            if (rec != null) {
+                // start from here again next time; step back over the tuple id
+                lastFound.set(rec.offset - LENGTH_TID);
+            }
+            return rec;
+        }
+        RecordPos findRecordInRange(final short targetId, final int from, final int to) {
             RecordPos rec = null;
-            for (int pos = 0; pos < dlen;) {
+            for (int pos = from; pos < to;) {
                 final short tupleID = ByteConversion.byteToShort(data, pos);
                 pos += LENGTH_TID;
                 if (ItemId.matches(tupleID, targetId)) {