Use lmv by default for insanity.sh.

update replay-dual.sh and replay-ost-single.sh

- Fix insanity.sh, still using the insanity-local.sh config by default.
- More cleanups in replay-single.sh. We need to generalize failures a
bit more, so we can choose a different mds each time, if more than one
is available.

Fix replay-single.sh and recovery-small.sh so they can run with the
default lmv config or "-f cfg/local.sh" config.

1)some fix for snap
2)add snap_add command

- test 4a of sanity-lmv.sh to check this case

- bunch of fixes to handle dir splitting over N < MDS nodes
  - fixes to MDS code to handle dir splitting over N < MDS nodes properly
  - mds_try_to_split() should ignore requests to split dir over 1 node
  - checks for NULL exports in LMV
  - checks for wrong -ERESTART handling: because of software errors
    it possible that lmv will loop forever. we need to assert on this.
  - one more sanity test 3c to test lfs dirstripe and activity on such a dirs

- cross-node mkdir() splits newly-created directory if requested
- remove_entries_from_orig_dir() accepts cross-node dir entries
- sanity check 3a tests splitting a directory with cross-node entries

- fixed error code leakage in filter_precreate() what led to endless loop
in the case of -ENOSPC due to real lack of space or inability to allocate new inode.

- added various error messages in error code pathes in lmv code.
- small changes in existent error messages.

2) remove and modify some test scripts according to the new lconf and lmc

- landed on HEAD

failover.sh renamed to mds.sh. This is a script for testing mds addiotion. Currently it does not use failover at all.

I haven't re-tested the AMD functionality since landing, but I did
test iozone and dbench to ensure sanity.

- lmv_iocontrol() which currently passes all commands to underlying mds's.
- some stuff in lproc_lmv.c, which shows now number of obds, number of active obds, etc.
- some cleanups, checks and error messages in lmv_obd.c.

- unlink for splitted dir has been implemented. it uses IT_UNLINK to check
  dir's emptiness and works as following: unlink requests comes to mds holding
  dir; mds recognizes splitted dir and issues LCK_EX with IT_UNLINK intent.
  each slave object is checked and locked. if all the slaves and master object
  are empty, then mds unlinks them and unlocks slave objects
- lmv_enqueue() handles splitted dir properly: issues given lock for each object
- lmv_unlink() handles splitted dir properly: unlinks each slave object
- mds_lock_slave_objs(), mds_unlock_slave_objs() and mds_unlink_slave_objs()
  have been introduced to implement splitted dir unlink
- mds_lock_and_check_slave() is IT_UNLINK handler on mds side
- mds_reint_unlink() sets MDS_MODE_REPLAY on "drop nlink" request in replay case
- mds_reint_unlink() recognizes MDS_MODE_DONT_LOCK and doesn't try to lock
  slave object being removed (it gets locked during earlier)
- minor cleanups in lmv to avoid needless debug messages
- sanity-lmv.sh has been added

Mostly fixes related to using failover MDSs along lmv. This is also related to active_ref handling, etc.

Also fixed the following:
* name of lproc entry where lustre upcall can be seen in error message if upcall is failed.
* removed --master option which was supposed to work along with lmv and now it is obsoleted.
* clanups and fixes both in lconf and lmc

- b1_4 is basically the parent of HEAD, because that's the direction
  that changes flow, as strange as that sounds.  so there's a
  HEAD_BASE tag which sits on b1_4

b_cmobd_merge should not be used for any further development or testing!

cMD_PRE_CMD_SMASH are the relevant tags.

The pre-smashing contents of HEAD will be committed by yury into b_cmd as a
merge of the two branches.  These will be stabilized on b_cmd and eventually
landed onto HEAD.

Added group lock tests. Those are not built by default.

b=2094, b=3138, b=3335, b=3274, b=3293, b=2862, b=3147, b=2350, b=3313
b=3265, b=3315, b=3301, b=3325, b=3329, b=3125, b=3098

r=Robert

Update the regex of conf-sanity.sh test 13 to parse mds uuid from xml file properly.

b=3284, b=3285, b=3043, b=3236, revert liblustre group changes

Stop fsx segfaults from re-using va_list arguments across two printf calls.
Also harden some of the fancy footwork involved in sizing an output buffer.
Also add some printf format checking and fix up some bugs it found.

b=3043, b=3258, b=3219, b=2410