Fix missing diff output on large diffs.

When there is a large diff chunk, match_lines() will skip running the
difflib.Differ.compare(). However this causes the following issues:

- It does not empty the `self.buf` buffer. This means that all future
  calls to match_lines() for that file will always be too large. So
  effectively no more diffs from the file get output.

- It outputs a debug message, but does not output anything to the
  side-by-side diff, so a user looking at the side-by-side diff may be
  misled into thinking the rest of the file has no differences.

We can fix these issue by falling back to a lazy line-by-line diff. This
produces suboptimal output, but it runs in linear O(n) time while
providing some form of output. We include a comment in the diff so the
user knows the following output is using a lazy diff algorithm.

diffoscope/diff.py

@@ -28,6 +28,7 @@ import threading
 import subprocess
 
 from difflib import Differ
+from itertools import zip_longest
 from multiprocessing.dummy import Queue
 
 from .tools import get_tool_name, tool_required
@@ -551,11 +552,9 @@ class SideBySideDiff:
         if len(l0) + len(l1) > 750:
             # difflib.Differ.compare is at least O(n^2), so don't call it if
             # our inputs are too large.
-            logger.debug(
-                "Not calling difflib.Differ.compare(x, y) with len(x) == %d and len(y) == %d",
-                len(l0),
-                len(l1),
-            )
+            yield "C", "Diff chunk too large, falling back to line-by-line diff ({} lines added, {} lines removed)".format(self.add_cpt, self.del_cpt)
+            for line0, line1 in zip_longest(l0, l1, fillvalue=""):
+                yield from self.yield_line(line0, line1)
             return
 
         saved_line = None