Remove all custody and VDF placeholders (56 lines less!). A few notes on the placeholder fields and logic:
* All placeholder fields were dummy fields that can easily be restored in the phase 1 hard fork.
* One special case to the above is `custody_bit_0_validator_indices` in `SlashableVoteData` which was renamed to `validator_indices`. Renaming it back is *not* a spec change because SSZ field names are [no longer part of the spec](a9328157a8 (diff-8d8fe480a35579c7be2f976d9b321216)).
* The placeholder logic was written using generalised functions (e.g. `bls_verify_multiple` vs `bls_verify`, and `indices(slashable_vote_data)` vs `slashable_vote_data.validator_indices`). This generality was unnecessary because it was not triggered when all custody bits were 0. This means we can simplify the logic without being inconsistent with phase 1.
Rationale:
* Keep phase 0 (likely far harder to deliver than phase 1) as clean as possible
* Focus on upgrade paths and incremental releases
* Custody is still under research—keep the design space open
Unlock decentralised pools using m-of-n threshold BLS signatures for both RANDAO and custody. We also simplify a bunch:
* Remove `randao_commitment` and `custody_commitment`
* Remove miscellaneous logic such as `repeat_hash`
(Side note: Dfinity seems to be working to reduce the communication complexity of BLS DKG (Distributed Key Generation) so validator pools should also benefit from that.)
Contents:
* Peg entries and exits to epoch boundaries
* Add a store of historical active index roots
* Mix it into the randomness
* Remove the delta hash chain
Note that the actual light client implementation is beyond the scope of the spec.
[Note to reviewers: verify that the invariant added in the PR is correct]
Question:
* Do we want to also only store epoch-boundary randao values? I don't think we use the epoch-intermediate ones anywhere.....
