Enable header peer catchup

When a new peer is selected to sync headers, it may start from far
behind, if there is a large backlog of bodies being downloaded
during fast sync. We were re-requesting the same headers from them
in an infinite loop (only resolved when we disconnect from the peer
and try again, after taming the backlog).

Author: carver

trinity/sync/common/chain.py

@@ -188,7 +188,17 @@ async def _sync(self, peer: HeaderRequestingPeer) -> None:
                         [header async for header in self._get_missing_tail(all_headers)]
                     )
                     if len(headers) == 0 and len(all_headers) > 0:
-                        self.logger.debug("All headers were redundant, requesting the next batch.")
+                        head = await self.wait(self.db.coro_get_canonical_head())
+                        start_at = max(
+                            all_headers[-1].block_number + 1,
+                            head.block_number - MAX_REORG_DEPTH
+                        )
+                        self.logger.debug(
+                            "All %d headers redundant, head at #%d, fetching from #%d",
+                            len(all_headers),
+                            head.block_number,
+                            start_at,
+                        )
                         continue
                 else:
                     headers = all_headers