availability-recovery optimizations

[sandreim]

Related to paritytech/polkadot#7411 . We should look further for optimizations for encoding/reconstruction to/from chunks. For example paritytech/reed-solomon-novelpoly#2 or taking advantage of latest SIMD instruction sets.

[burdges]

cc #598

[sandreim]

Adding https://github.com/AndersTrier/reed-solomon-simd to the options to explore

[ordian]

This will be a breaking change to the format, but it might be worth it - can be coordinated with node_features.

Also, while at it, we can also finally fix the merkle commitment to be a binary tree instead of complicated hexary trie.
All of this is done in https://github.com/paritytech/erasure-coding and the benches show a decent gain in encoding as all: https://gist.github.com/ordian/0af2822e20bf905d53410a48dc122fd0

[burdges]

Ahh nice, it switches to reed-solomon-simd too. We observe a speed up like the one he claims? :)