Raw tables: Support rest column

[simolus3]

This adds support for the Rest column described in the Raw Tables roadmap.

When the Rest column is used, we create a set of all other columns matched in a statement. During the sync process, we then match over the received row and build a JSON object containing values from the source row that aren't matched by a column in the raw table.

Having a rest column in the raw table can help with some migrations that add columns, because synced values would be available in that table without having to re-sync.

[rkistner]

Just for my own curiosity - are the rest fields parsed and re-serialized, or do we just keep the serialized form as-is?

[simolus3]

are the rest fields parsed and re-serialized, or do we just keep the serialized form as-is

We've already been deserializing each oplog row as a JSON object to be able to extract individual columns. Building the value for rest reuses that parsed object instead of deserializing again, if that's what you mean?

We deserialize into a Map<String, serde_json::Value> - that allocates for string values where we could get away with an &str reference if the string doesn't contain escape characters. So there is some potential for optimizations here, but I don't expect this PR to make matters worse.