Fix period tail in get_eth1_vote

[djrtwo]

This is a cleanup of get_eth1_vote inspired by @paulhauner's PR #1463 and subsequent conversation.

The previous version was incorrectly assuming that there was an eth1_data_vote per slot, but this is not the case when 100% of proposers don't get blocks into the chain. Instead the period tail should be based upon the actual slot.

This PR simplifies the logic so that in the early slots of a voting period, you only consider new_eth1_data when deciding your vote, but in the period tail to encourage settling on a vote, you consider all_eth1_data.

[protolambda]

LGTM