Update trie spec

[lamafab]

From Emeric:

There is a change in the header: two new variants. And so two new node types
Not exactly new node type, existing type but with hash of value stored instead of value
I am also not sure to what extent proof were specified. This does change a bit compact proof (which are used by cumulus and one rpc), but I don't think this is part of the spec
Out of trie scope, it changes one host function to chose which trie/state version should be applied, that is an additional parameter on 'storage_root'.

On the trie crate it is paritytech/trie#142 which allow storing value out of the node.
The encoding scheme is compatible with the existing one as long as it is only reading (the header allows both old and new behavior, with value always in a trie node or with value bigger than a hash stored as an external node).
In substrate it is paritytech/substrate#9732, the same codec code change is used and there is an additional 'state_version' (the new parameter of the host function) in the that indicate if the new behavior should be use when modifying the state.
There is a third PR that focus on runtime change to run a migration (paritytech/substrate#10073).

[lamafab]

Implemented in #478