See comment. The checkpoint parsing code does not currently parse chained applyOps for the replicated metadata checkpoint. The write is unlikely to be split because it requires 80k collection changes in a single checkpoint, but it falls within the upper bound of allowable collections.

See also https://github.com/10gen/mongo/pull/65018#discussion_r4351536820. This should go away if we properly handle checkpoints split across entries, but the main point is that we do not allocate intervals without some kind of eviction mechanism for partial/incomplete checkpoints. Recall that upon seeing the next checkpoint, we will evict the previous intervals.