pallet-mmr: handle forks without collisions in offchain storage (#11594)

* pallet-mmr: fix some typos

* pallet-mmr: make the MMR resilient to chain forks

* pallet-mmr: get hash for block that added node

* beefy-mmr: add debug logging

* add explanatory comment

* account for block offset of pallet activation

* add support for finding all nodes added by leaf

* minor improvements

* add helper to return all nodes added to mmr with a leaf append

* simplify leaf_node_index_to_leaf_index

summing the (shifted) differences in peak positions adds up to the (shifted) final position, so
don't need to fold over positions.

* dead fish: this also doesn't work

The idea was to keep a rolling window of `(parent_hash, pos)` leaf
entries in the offchain db, with the window matching the one
that provides `block_num -> block_hash` mappings in `frame_system`.

Once a leaf exits the window it would be "canonicalized" by switching
its offchain db key from `(parent_hash, pos)` to simple `pos`.

This doesn't work however because there's no way to get leaf contents
from offchain db while in runtime context.. so no way to get+clear+set
leaf to change its key in offchain db.

Ideas:
1. move the "canonicalization" logic to offchain worker
2. enhance IndexingApi with "offchain::move(old_key, new_key)"
   This is weird, but correct, deterministic and safe AFAICT, so
   it could be exposed to runtime.

* simplify rightmost_leaf_node_index_from_pos

* minor fix

* move leaf canonicalization to offchain worker

* move storage related code to storage.rs

* on offchain reads use canonic key for old leaves

* fix offchain worker write using canon key

* fix pallet-mmr tests

* add documentation and fix logging

* add offchain mmr canonicalization test

* test canon + generate + verify

* fix pallet-beefy-mmr tests

* implement review suggestions

* improve test

* pallet-mmr: add offchain pruning of forks

* pallet-mmr: improve offchain pruning

Instead of keeping pruning map as single blob in offchain db,
keep individual parent-hash lists with block-num identifier as part
of the offchain key.

Signed-off-by: acatangiu <[email protected]>

* pallet-mmr: improve MMRStore<OffchainStorage>::get()

Do the math and retrieve node using correct (canon or non-canon)
offchain db key, instead of blindly looking in both canon and non-canon
offchain db locations for each node.

Still fallback on looking at both if for any reason it's not where
expected.

Signed-off-by: acatangiu <[email protected]>

* pallet-mmr: storage: improve logs

* fix tests: correctly persist overlay

runtime indexing API works on overlay, whereas offchain context
bypasses overlay, so for loops > canon-window, canon would fail.

* pallet-mmr: fix numeric typo in test

* add comment around LeafData requirements

Signed-off-by: acatangiu <[email protected]>

Co-authored-by: Robert Hambrock <[email protected]>