The number of possible lock list entries for a thread is
LOCK_CHILDCOUNT, and each entry can record up to LOCK_NCHILDREN locks.
When iterating over the list of sleep locks held by a thread, assert
that the loop index is less than LOCK_CHILDCOUNT * LOCK_NCHILDREN rather
than witness_count. The latter determines the size of the lock order
matrix and is thus unrelated, so the previous assertion was incorrect.

None, but this change just weakens an existing assertion. We ran into this at Isilon because we have threads that can acquire a large number (> 2000) of shared locks.