Tree based spec SSZ implementation - remerkleable

[protolambda]

This mostly just deletes the old SSZ implementation from the spec.
And then if we link in this library: https://github.com/protolambda/remerkleable, it passes all spec tests.

The library is different from status quo in the following ways:

• data-sharing at the core: every type is a view over a binary merkle tree.
• supports typing over partial backings: just provide the branches of the tree that matter to the reads/writes on the object, and it works.
• hash-function doesn't even show up in initial profiling, everything is cached in place
• data accesses are more expensive, as it creates temporary views over subtrees. E.g. a validator view for each registry entry to get active validator indices.
  • Quadratic-complexity state accesses slow down things, even more than a minimal-configuration state hash-tree-root. Naively calling a function for every validator, that then calls a function for every validator, is a big slowdown. Normal eth2 clients wouldn't do this, but the spec does things like this in reward-computation.
• For normal use (not the quadratic problem as described above), mainnet config is much more manageable for testing, state-hashes are fully cached.
• Typing/functionality extensions and improvements over status quo spec implementation
• Deserialization now fully supported.
• Lots of other SSZ feature goodies from remerkleable to use in tooling.

And compared to py-ssz:

• More readable types, pretty much the same syntax as status-quo spec.
• Caching and data-sharing everything, with access to the binary tree (good for loading typed merkle proofs at 0 cost). As opposed to pyrsistent, the data-sharing library that py-ssz uses.
• (de)serialization should be slightly faster, passing an stream around to read/write from.

PS: pymerkle is already taken as a name, pymerkles was a working name. Suggestions for a good name are welcome. Something that covers ssz / merkle partials better. Decided on a new name: remerkleable! A mix-up of "remarkable"/"re-merkle-able".

In addition, the new cached hash-tree-roots are very useful for memoization of spec functions, bring big performance improvements to the pyspec code.

[protolambda]

Need Python 3.8 for remerkleable typing features (as well as changes in phase 1 PR). Deposit tests are failing with 3.8 though, since vyper fails parsing when using the 3.8 AST code. Need to backport the PR fromt the vyper repo that fixed that.

[protolambda]

Hmm pypi db index is slow, CI can't find v0.1.9 I just published. Will re-run after giving it some time to catch up

[protolambda]

This PR is ready for review.

Some final notes:

• I checked the ssz_static generator output against that of the previous ssz solution, and it matched, except for some files that had differently generated random values. The generator is not perfect.
• Although the memoization is a big performance gain, the memoization is backed by just dict now. I'll swap in a proper memoization cache in another PR.

[djrtwo]

Looks good. Minor merge conflict with build_spec.py

[protolambda]

Just implemented ByteList, testing and preparing a new remerkleable release to include here.

[protolambda]

Reduced it all to one single commit 👍

[djrtwo]

looks good! ready to merge