Implement wait-free doppelganger detection

[zah]

Upon start-up, Nimbus can check whether the local slashing protection database suggests that the node was just restarted (by looking for records from the epoch immediately preceding the current one). Under such conditions, it's safer to start attesting immediately. We can call this mode --doppelganger-detection:auto and it will be suitable as a new default.

[tersec]

prysmaticlabs/prysm#7985 (comment) describes one approach to this.

[mratsim]

For history: Discord Message from 2021-04-13

re: Doppelganger Detection. Looking at the issues in LH:

• Enable Doppleganger Detection to reduce the likelihood of slashing sigp/lighthouse#2069
• [Merged by Bors] - Doppelganger detection sigp/lighthouse#2230

I think there are 2 scenarios where we can skip waiting:

1. The validator attached had emitted 2 attestations for the previous 2 epochs and they are in the slashing DB.
   As an initial fast check we can check that the count of attestations in the slashing DB is equal to 2x number of attached validators. (so that we have global handling instead of per-validator wait.)
2. Genesis.

Otherwise if there is a gap I don't think we can say for sure that there isn't another validator.

Now the lighthouse issue brings up a fun point. If everyone has the detection turn on, and restart at the same time, the chain might stall.

Now that we use sqlite as a DB (and not a kvstore), we can easily do range check for point 1. And point 2, was closed after we launched I-don't-remember-which testnet

[arnetheduck]

Good thing we have 3s restarts now then :)