Ingest tables are in-memory only, meaning they don't include checkpoint information that needs to be persisted in the metadata. However, the current approach still inserts an entry for the ingest table into the local metadata.

As an alternative, we can infer the metadata for an ingest table based on the metadata of the corresponding layered table. When opening an ingest table, we can first look up the metadata for the layered table, then derive the ingest table’s metadata using an empty checkpoint. This inferred metadata can be used to open an empty ingest btree.

By adopting this approach, we can avoid persisting the metadata for ingest btrees and eliminate the need to create missing ingest btrees when loading a new checkpoint.