ChangeNotesParser: Avoid lots of FooterLine allocations

The implementation of RevCommit#getFooterLines(String) always calls
getFooterLines(), which creates a new list filled with newly-parsed
FooterLine objects. This seems wasteful, especially considering in
ChangeNotes we call getFooterLines 10 or so times per commit.

Add a subclass of RevWalk that caches the result of getFooterLines in
a multimap. I don't claim that this is the most efficient
implementation possible, but it does give us the abstraction we need
to optimize more later.

Some tests were failing in a previous iteration of this change when I
didn't implement case-insensitive behavior. Normalize
ChangeNotesParserTest to use exactly-matching case in labels, except
for a single test method that explicitly tests case-folding behavior.

Change-Id: I836afa43cde2c1d8df04286664a3c6cc0fac57a9

gerrit-server/src/main/java/com/google/gerrit/server/notedb/DraftCommentNotes.java

@@ -35,7 +35,6 @@ import org.eclipse.jgit.lib.ObjectReader;
 import org.eclipse.jgit.lib.Repository;
 import org.eclipse.jgit.notes.NoteMap;
 import org.eclipse.jgit.revwalk.RevCommit;
-import org.eclipse.jgit.revwalk.RevWalk;
 
 import java.io.IOException;
 
@@ -169,7 +168,7 @@ public class DraftCommentNotes extends AbstractChangeNotes<DraftCommentNotes> {
       }
       ObjectId draftsId = newState.getDraftIds().get(author);
       repo.scanForRepoChanges();
-      return LoadHandle.create(new RevWalk(repo), draftsId);
+      return LoadHandle.create(ChangeNotesCommit.newRevWalk(repo), draftsId);
     } catch (NoSuchChangeException e) {
       return super.openHandle(repo);
     } catch (OrmException | ConfigInvalidException e) {