Change BLS signature/public key aggregation to enable better caching

[Stebalien]

Right now we're not taking advantage of the bitfield mask when aggregating public keys. I.e., each time we aggregate public keys we use an all-ones bitfield and cannot re-use any of the results of the aggregation.

Instead, we should:

1. Use the full power table when aggregating public keys, using the bitfield to select the active keys. This will cost a bit more up-front, but we'll only pay that cost once per instance.
2. Cache the coeff*pubkey+pubkey for every public key for the rest of the instance (cache https://github.com/drand/kyber/blob/94dae51d79b4b0c2d2a9b9cc382b864cf3537783/sign/bdn/bdn.go#L183-L185).

This is the one N^2 cryptographic operation we do, so it should make F3 scale significantly better.

[Stebalien]

drand/kyber#61

[Stebalien]

In GPBFT, my plan is to change the Verifier interface as follows:

type Aggregate interface {
	// Aggregates signatures from a participants.
	Aggregate(ids []int, sigs [][]byte) ([]byte, error)
	// VerifyAggregate verifies an aggregate signature.
	// Implementations must be safe for concurrent use.
	VerifyAggregate(payload, aggSig []byte, signers []uint64) error
}

type Verifier interface {
	// Verifies a signature for the given public key.
	// Implementations must be safe for concurrent use.
	Verify(pubKey PubKey, msg, sig []byte) error
	// Return an Aggregate that can aggregate and verify aggregate signatures
	// made by the given public keys.
	Aggregate(pubKeys []PubKey) Aggregate
}

One downside to this approach is that we'll pay the cost of pre-computing the coefficients for all keys in the power table when validating finality certificate signatures where the current method only has to compute the coefficients for the keys actually used. But... I don't think that's going to be a huge deal and I want to keep this simple for now.

[Stebalien]

Ew. So... validation makes this a bit annoying. We're going to have to cache this as part of the committee, possibly put it in gpbft.PowerTable (for validation). Probably the latter, TBH.

[Stebalien]

Hm. PowerTable is the right place, but that's probably going to cause a bunch of issues. I'll probably have to add it as an additional param and cache it in the *committee object.