Merging heterogeneous fragments into a single lance dataset

[artemru]

A practical question: let's say, several fragments have been written in distributed way (following https://lancedb.github.io/lance/guide/distributed_write/#write-new-data), but for some reasons (different processing paths, dirty inputs, ...), not all of fragments have the same schema. Typically, some of fragments have some missing columns.

What is the easiest way to create a lance dataset from those fragments with an unified schema ? Ideally, without rewriting anormal fragments, but just changing their metadata or so.