revlog: avoid large yields in group()

Split large yields so that the downstream consumer (chunkbuffer) will avoid
some pain when reading ahead.

mercurial/revlog.py

@@ -1096,6 +1096,13 b' class revlog(object):'
                 d = self.revdiff(a, b)
             yield changegroup.chunkheader(len(meta) + len(d))
             yield meta
+            if len(d) > 2**20:
+                pos = 0
+                while pos < len(d):
+                    pos2 = pos + 2 ** 18
+                    yield d[pos:pos2]
+                    pos = pos2
+            else:
             yield d
 
         yield changegroup.closechunk()