Inner hash of value with new nodes

[cheme]

This PR aims at replacing #134.
It achieves the same goals but always use a new db node when inner hashing is use.
This makes code simpler and can be interesting with rc dbs.

Note that compact proof and compact proof with value iterator did change a bit:

• compact proof: hashed value are following the node (node is switch to a standard node with 0 length value), which is prefixed with a header dead value.
• compact proof with value iter: hashed value when ignored are replace similarily, just no value node are added afterward.

At this point review can be started as all remaining TODOs are minor.
Codec in substrate_like.rs is the same as the one which will be ported in substrate except for the default threshold size (set to 1 here as most of test only use small values).

TODOs

• crate versioning

• write a test for value removal with prefix (check the value node prefix is correct) + a test that insert twice in triedbmut

• 63e1651 is related to a possible migration optimisation in state machine, does not really have to be in this PR.

• consider running rust format (little care was used here)

[cheme]

b7238ef does remove the hybrid specific compact proof encoding.
@thiolliere , ESCAPE_HEADER was not remove, it is still needed in trie_codec.rs (when header is present we got attached value after the hash). I cannot remove it because the codec is not writing value length when encoding a hashed value so we need to know if value is in proof or not.

[gui1117]

looks good to me, I didn't understood deeply all algorithm, especially the proof in trie-db/src/proof/

But all changes made sense to me.