Recursive read-resource should not fail if an expected address type disappears in the middle of execution

The ManagementResourceRegistration tree is not cloned upon modification the way Resource tree is. That means read ops executing concurrently with ops that modify the MRR tree can have the tree change underneath them.

This can affect recursive read-resource ops, as they calculate possible child addresses and then later they process those addresses, failing if an MRR is missing. A failure only makes sense if the user did something wrong or, perhaps, if it's clear that the code is broken. But here there's a perfectly valid reason for the MRR to be missing, so there should be no failure.

Not an urgent thing; I saw this quite a while ago when investigating a different issue and quickly worked up a fix which has now been sitting around gathering dust.