copy: speedup by making overlap the default

Since commit 62e2f5944b, aio prevents races on the length
of a transaction logfile.

Thefore, we can safely enable IO parallelism at writes fired off
by copy.

The old behaviour was a serious IO bottleneck.

kernel/mars_copy.c

@@ -32,7 +32,7 @@
 
 #include "mars_copy.h"
 
-int mars_copy_overlap = 0;
+int mars_copy_overlap = 1;
 EXPORT_SYMBOL_GPL(mars_copy_overlap);
 
 int mars_copy_read_prio = MARS_PRIO_NORMAL;