Add bank-timestamp design proposal

[CriesofCarrots]

Problem

Bank::unix_timestamps() are drifting further and further from real-world time.

Summary of Changes

Propose a method for correcting this drift using the validator-timestamp oracle

[mvines]

I think the major part that we're missing here is how the drift is actually corrected:

1. If the bank clock is ahead of wallclock, then at the end of the epoch the bank clock is perhaps incremented by 1 (the smallest amount larger than 0)
2. If the bank clock is behind of wallclock (like it is currently), how quickly do we warp ahead? I feel like we probably shouldn't warp ahead too much. In the case of a cabal of validators attempting influence over the clock, the only way to reduce their impact will be to delegate stake away from them. This will take days/weeks. So perhaps we only warp head by at most 1 epoch of wallclock? But if we limit to this, will we reach close to realtime by the end of the year?

[CriesofCarrots]

I'll add a section on this.

1. I was thinking we wouldn't ever correct the bank clock backwards, so if the bank clock is ahead of wallclock, it isn't adjusted at all.
2. Assuming that the real average slot duration remains similar to that over the past 40M blocks, one additional epoch's-worth of correction will not be enough to get us synced by the end of the year. According to my math, we'll need to correct approximately 2.5 additional epochs of wallclock each epoch to reach real time by the end of the year.

[CriesofCarrots]

Closing in lieu of a simpler approach: #12737
Will re-open if needed