In order to improve the speed of landed cost validation, we defer the
creation of AML at the very end of the process. This way, we don't
suffer from various performance issues caused by:
- cache invalidation (use of `with_context`)
- recomputations

opw-1824661