We had a smiliar approach. We had one xml file per table to have a good overview over every table and its columns. But when we needed to refactor(e.g. add new columns, move some data, add new FKs) our db, it was revealed, that this approach comes with some problems. The possibility to see what belongs to one table and the execution order are in conflict with each other. My best(but still dirty) solution was to start a single new xml file named migration_X.xml with all changeSets to add columns, copy some data, add new constraints or FKs.
If I want to see the definition of one table, I generate the DDL.