branchcache: improve speed relative to the amount of heads

Updating the branch cache is quadratic to the amount of heads in the
repository. One consequence of this was that cloning a pathological
repository with 10,000 heads (and nothing else) took hours of CPU
time.

This patch makes one of the inner loop much faster, by removing a
changectx instantiation, and removes another entirely in cases where
there are no candidate branch heads which descend from other branch
heads.

mercurial/localrepo.py

@@ -509,15 +509,17 b' class localrepository(repo.repository):'
             bheads.extend(newnodes)
             if len(bheads) <= 1:
                 continue
+            bheads = sorted(bheads, key=lambda x: self[x].rev())
             # starting from tip means fewer passes over reachable
             while newnodes:
                 latest = newnodes.pop()
                 if latest not in bheads:
                     continue
-                minbhrev = self[min([self[bh].rev() for bh in bheads])].node()
+                minbhrev = self[bheads[0]].node()
                 reachable = self.changelog.reachable(latest, minbhrev)
                 reachable.remove(latest)
-                bheads = [b for b in bheads if b not in reachable]
+                if reachable:
+                    bheads = [b for b in bheads if b not in reachable]
             partial[branch] = bheads
 
     def lookup(self, key):