When filesink-based materializations calculate paths for files, they use the x-collection-name from the source as the identifier in the target path, along the lines of bucket/prefix/v00001/v00000001/collection_name.
Broadly, the platform supports a generalized concept of a schema associated with a collection, which is standardized as the second-to-last part of the collection name. We could use this as part of the file path as well, so that it becomes bucket/prefix/v00001/v00000001/schema_name/collection_name.
This would be beneficial in cases where there is more than one source binding with the same final path component, but differing schemas.