* Bring forward changes to withdrawability from phase 1
* The `WITHDRAWABLE` flag is removed; instead, a validator's withdrawability is determined through the `withdrawable_epoch` field (renamed and re-purposed from `withdrawal_epoch` which was not used)
* When a validator passes through the withdrawal queue, the `prepare_validator_for_withdrawal` function does not let them withdraw immediately; instead, they have to wait `MIN_VALIDATOR_WITHDRAWAL_EPOCHS`. This extra minimum delay serves no value in phase 0, but is crucial for phase 1 as the period between a validator passing through the queue and the validator being eligible to withdraw is where proof of custody challenges can come in; adding it in phase 0 is only half a line of code so easier to add it now.
* If a validator is penalized, they are no longer subject to the exit queue; instead, their `withdrawable_epoch` is set `LATEST_PENALIZED_EXIT_LENGTH` into the future and this is used to determine when the validator can withdraw
* Changes the eligibility condition for a transfer to use the `withdrawable_epoch`
1. Use `+` to concatenate the merkle roots in `hash` function.
2. Fix `pad_to_power_of_2`: padding with `[b'\x00' * SHARD_BLOCK_SIZE]`,
not `[SHARD_BLOCK_SIZE]`.
Reasons to use `merkle_hash` instead of `hash` for containers:
1) **Smaller witnesses**: `BeaconState` is a somewhat wide container (26 fields as of now, likely 30+ in phase 2). With concatenation and plain concatenation the size of the Merkle witnesses for the top level are ~32 bytes per field element.
2) **Faster incremental hashing**
3) **Consistency**: Consistent with `merkle_hash` for lists/vectors.
Adds the crosslink committee to the fork choice rule. This is useful because it means that even if a proposal committee is byzantine and attempts to prevent a crosslink via a "balance attack" (alternating between chain A and chain B being the canonical chain), the crosslink committee can force the equilibrium to flip to one side or the other.