Change if statements in get_crosslink_committes

[paulhauner]

Fixes #639.

[JustinDrake]

Why doesn't get_previous_epoch(state) always return get_current_epoch(state) - 1?

[djrtwo]

^ remnant from when we started epoch at 0 to prevent underflow.
We should resolve #626 before we modify get_previous_epoch. Seems like people want to see epochs/slots to start at 0

[JustinDrake]

I see three ways to prevent underflows:

1. Add a bunch of subtle and crufty if special cases across the codebase. We already tried this and In hindsight it feels like the wrong way forward.
2. As suggested in The integer debate #626, use int64 (as opposed to uint64) for slot and epoch and start with GENESIS_SLOT = 0. Definitely better than 1) but maybe not quite as clean as 2).
3. Have GENESIS_SLOT be large enough. It was my mistake to not realise that 2**63 would cause complications with Java and it's unfortunate there's no obvious choice. My latest proposal is 2**n * SECONDS_PER_DAY/SECONDS_PER_SLOT where n is the smallest sufficiently large exponent.

[djrtwo]

@paulhauner #655 addresses this immediate issue. I'm combing the spec to make sure that the initial previous epoch being before genesis_epoch doesn't open up any unexpected new bugs