Migrate chunks on ingestion

[emilk]

When loading a chunk containing old data, we want to migrate it by converting its contents to the latest format.
This probably mean using the old generated Rust component/datatype deserializer, having some custom transform run on it, and then writing it back with the new serializer. For archetypes it may require loading N columns and outputting M new ones.

For this to work we first need

• Add versioning info in the arrow metadata #8748

What crate does this migration work? re_sorbet seems like a likely candidate, if we think it's fine that re_sorbet depend on re_types.

Initially we only need to support loading the two most recent version of data, not all previous versions.

In most cases this can be zero-copy.

Alternatives

We provide a migration tool, e.g. rerun migrate foo.rrd that updates an .rrd to the latest version. The downside is that this requires more work for users, and also won't work for on-wire data (e.g. coming from an older storage-node over gRPC).

Progress

• We currently migrate the names of some archetypes and field names (Deprecate SeriesLine/SeriesPoint/Scalar in favor of SeriesLines/SeriesPoints/Scalars #9338).