We only use wp_value in the end if the resulting goal is yet
another wp. Otherwise we may not be able to do a final view
shift (as observed by Ralf).

It now also reshapes expressions as values for contexts that need
values such as AppECtx.

This is shorter and more consistent with naming elsewhere.

This makes the proofs a lot nicer. I tried factoring out some properties
about finite maps, but this was not entirely satisfactory. It would be nice
to generalize from_heap_None and heap_singleton_inv_l somehow.

This works better with class interference.

start the heap construction: load from a heap is done... store, CAS, and singleton (mapsto) to be done

Also, make our redefinition of done more robust under different
orders of Importing modules.