Slashing penalty calculation change

[vbuterin]

If the exit queue is very long, then a validator may take many months to exit. With the code as currently written, however, self-slashing is a potentially lucrative route to get one's money out faster, because one can exit in 36 days.

This PR changes it so that slashing can only extend your withdrawal time, not contract it. Also, instead of the slashed balances used to calculate one's slashing penalty being those in [withdrawal - 54 days ... withdrawal - 18 days], we now run the penalization algorithm once every 36 days that a validator is slashed but not withdrawn, so that it covers the 36-day period where the validator was actually slashed. It also moves the minimum slashing penalty to the slash_validator function so that it is only applied once.

[djrtwo]

I'm fixing tests as I review

[djrtwo]

Looks good!
Fixed and added some tests

[JustinDrake]

I've revamped and simplified the slashing logic:

1. Avoid the ever-growing sum in slashed_balances. Instead, simply store the per-epoch sum. This is less confusing, allows for simpler logic (simply do sum(slashed_balances)), and removes the long-term risk of slashed_balances overflow.
2. Rename slashed_balances to slashings. Shorter, consistent with BeaconState group label, just as descriptive.
3. whistleblowing_reward => whistleblower_reward (for consistency with proposer_reward)

[djrtwo]

reasonable change to the slashing vector. That used to be the most likely thing in the state to overflow