sohuldn't the type of this be NodeIndex?

In #8407 and #8408, we forked slab, the crate that underpins our node arena, to add support for generational indices: each index now tracks the position into the memory vector (one u32) as well as a monotonically increasing counter (another u32), which gets bumped every time that slot in the memory vector is reused.
This prevents us from hitting the ABA problem:

• You allocate a new node in slot 5, getting index 5.
• You remove the node in slot 5, it goes back on the free list.
• You allocate a new node, slot 5 gets reused for a new, unrelated node.
• Some other part of our code still holds the old index 5, thinking it points to the original item.
  It reads or writes through that stale index, silently accessing the wrong node.

I improved the docs here to explain how they relate to the index, and the documentation of the node index itself.
Without generational indices, we would have to abort GCing whenever the revision id changes, but that would be way too aggressive.

remind me what node_generation is?
Consider documenting