VM reliability

The vm in 2.4.17pre4aa1 should be rock solid now. The design avoided all the superfluous parts while maintaining the interesting parts of the early 2.4 VM (like all the delayed writes/kiobuf pinning logic/swap_out walking).
Finally in the last weeks also the google testcase is been fixed. The google testcase was really reproducible trivially by just allocating lots of giga of ram, and then starting doing intensive I/O, so Oracle and all the other applications allocating lots of memory and doing lots of I/O were heavily affected too.
The new 2.4.15aa1 VM code is just in production.