kernel - More readdirplus deadlock avoidance work

kernel - More readdirplus deadlock avoidance work

* Add cache_nlookup_nonblock() and nfs_nget_nonblock()

* Adjust the readdirplus code to use the new functions.   Basically
  there are too many locks being held by callers and then readdirplus
  goes and tries to acquire (and lock) nodes for the directory elements
  on top of that, leading to deadlocks.

  Attempts to order the locks have failed so now we just use a
  non-blocking approach since the readdirplus entries are all
  advisory anyway.