Track lineage when merging/splitting chunks

[teh-cmc]

We currently generate brand new ChunkIds whenever we merge/split chunks (e.g. during JIT compaction), because it's the easy way to prevent collisions with minimal effort, but that comes at the cost of losing track of all of the Chunk's lineage.

We should at the very least keep track of these things in the chunk's metadata, e.g.:

* rerun:parents: [chunk_xxx, chunk_yyy]