Improve attribute data structure for gtf_scan

[jdcla]

Currently gtf_scan returns the attribute column as a list of structs that each have two fields ('key' and 'value'). Unfortunately, this setup makes it hard to unnest this data in a tabular format. Having the data either fully unnested or in a more helpful structure would be more convenient.

For example, it might make more sense to have the attributes listed in a single struct, where the field names equal the 'key' values, making it possible to retrieve an item using polars directly ('e.g. pl.col('attributes').struct.field('gene_id')). Now, the list has to be iterated and searched for a specific key, denoting an expensive operation.

Lastly, when multiple values for a specific key (e.g. tag) are present (listed as a list ccds, basic), these are currently divided in their own struct tuple.