Improve Pixel Alignment iteration

Iterating over the `pixel_mapping` in a `PixelAlignment` is currently quite awkward, and appears repeated in separate places such as crossmatching and joining. Some way of improving this would be useful, maybe by making helper methods in the `PixelAlignment` class or by changing away from the data frame representation.

Start by locating the PixelAlignment class and its pixel_mapping representation, then find the repeated iteration in crossmatching and joining. Compare whether shared helper methods or a different representation best reduces the duplication. Done means iteration is simpler and consistently handled in both areas.

Written by the indexing model from the issue text.