Multiple column family support

[bobvawter]

https://www.cockroachlabs.com/docs/v22.1/changefeeds-on-tables-with-column-families

• [ ] Separate a MutationKey type from Mutation (code cleanup to create a comparable type, no functional changes).
• [ ] Stub out column family plumbing in MutationKey.
• [ ] Introduce a v2 staging table schema, similar to the newer checkpoint table, that can persist and retrieve multiple CF entries for a row at the same timestamp.
• [ ] Support for sparse payloads in the apply package (i.e. drive sql generation from fields in payload rather than using all columns in target schema).
• [ ] Support decoding CF topic entries in incoming payloads.

[BramGruneir]

What are you thinking here?

It seems like unnecessary complexity.

[bobvawter]

Mutations are currently identified by (table, pk, time). We'll need the ability to track a partition (column family) name as well. As a preparatory step, it's probably worthwhile to extract a MutationKey type from types.Mutation to ensure we can identify all features which depend upon the identity of a mutation. Once there's an API concept that encapsulates a mutation's identity, we can expand it to support column families or other dimensions we haven't previously considered.