Handle incomplete data around restarts of beacons

[aimxhaisse]

Currently the watcher depends on the liveness API to know if attestations are missed:

https://ethereum.github.io/beacon-APIs/#/Validator/getLiveness

A beacon node that was recently started or suffered a network partition may indicate that a validator is not live when it actually is.

We sometimes end up reporting missed attestations at around beacon restart due to this, one improvement here would be to wait at start that the beacon has at least 2 epochs of uptime. Unsure how we can infer this from the beacon API but there may be ways.