shrink-revlog: improve performance: use changegroup instead of revisions
Before:
real 0m23.971s
Now:
real 0m4.229s
The only case where the output would be different is if the newer hg was using
a different diff algorithm than used originally.
% init
adding a
% fold in the middle of the queue
popping p3
popping p2
now at: p1
--- a/a
+++ b/a
@@ -1,1 +1,2 @@
a
+a
p1
p3
--- a/a
+++ b/a
@@ -1,1 +1,3 @@
a
+a
+b
% fold with local changes
abort: local changes found, refresh first
--- a/a
+++ b/a
@@ -1,1 +1,3 @@
a
+a
+b