Payout function should refund

[kianenigma]

Current payout function of staking assumes that any payout pays exactly T::MaxNominationPerValidator accounts, not to mention that it also assumes the worse case of account target as well: Staked.

We should make this refund along two axes:

1. If we pay X nominators, we should take this into account.
2. Each payout variant should have its own weight benchmark, and we must add them up. So the function will still use the upfront worse case weight, but will refund quite a bit along the way.

Once we have this, we can bump the MaxNominationPerValidator to a much larger number safely, effectively nilling the chance of oversubscribed nominators.

related paritytech/polkadot#2637

[shawntabrizi]

Each payout variant should have its own weight benchmark, and we must add them up. So the function will still use the upfront worse case weight, but will refund quite a bit along the way.

I think this is probably overkill. Just track how many below the max payouts we actually made and make a refund there?

[kianenigma]

Arguably yes, the former is much more important.

[kianenigma]

@emostov wanna try this one to show us what you've got? :P