tdb: Do not return errors from tdb_repack() in the tail of tdb_transaction_commit()
The call to tdb_repack() inside tdb_transaction_commit()
is an optimization, not part of the transaction itself,
so failing due to lock or other errors isn't a fatal error
that should cause the caller to think the transaction was
a failure by returning -1.
The tdb transaction itself has finished and been committed
onto stable storage via fsync and all locks released at the
point tdb_repack() is called.
tdb_repack() is only called here as it's a convenient point
to attempt to reduce tdb fragmentation without having to add
a timer call to repack in all users of tdb.
This causes lock ordering issues in Samba, showing up as: