This class whose name is horrible and purpose is arbitrary seems to be a
leftover of some experiment with ch2o, a long time a ago.

That way type class search becomes more predictable.

Lifting lemmas : make the initial state appear under an existential under the view shift. This is still sound, but slightly more expressive.

This may save the need to seal off some stuff.

New lemma : cmra_update_valid0. This let us prove a FP update using the additionnal hypothesis that the source is valid at step 0.

In particular, it no longer uses set_solver (which made it often slow
or diverge) but a more specific lemma about subseteq.

In noticed in Amin's development that importing the proof mode often
turns length into String.length. The weird thing is that before importing
the proof mode, it refers to List.length, and when importing just the
proof mode, it refers to List.length too. However, in some combinations of
imports, it seems to result in it refering to String.length...