Overhaul data reloading

Overhaul data reloading

The data model change left the reload feature in a miserable state
causing facets of mutable elements to be lost, changes introduced in
read-only origins to be ignored, and so on. Simplify the
implementation to just save, wipe and re-read all persistent data as
if the engine were starting initially. Also add a basic check to the
reload test to make sure a mutable element still has facets after
reloading, so that we reduce the risk of future regression. Include
a bit more verbose logging around when and what files are read at
load/reload time.