High performance ECDH

[dcousens]

ecurve is slow, but battle tested.

Related #350
#623
#459
#508
#407

[dcousens]

bip32 will be using the secp256k1 library on release (soon)... but I haven't yet finalized an ECC API that I am happy with in bitcoinjs-lib itself.

We want to move to secp256k1 before Christmas 😄 .

I like the idea of keeping the .sign() object function so that users can swap in whatever library they need with abstractions like TransactionBuilder etc.
But, in ECPair, with getAddress gone, the utility of the abtraction has dropped to near zero.

I don't want to pass a Buffer(Length: 32) private key to TransactionBuilder.prototype.sign though... so what should we do?

An idea was to leave it entirely up to the user:

let keyPair = ECPair.makeRandom()

// ...
txb.sign(i, SIGHASH_ALL, function (hash) {
   return keyPair.sign(hash) // returns 64 bytes R|S encoded
})

Where we can offer a easy-sauce secp256k1 compatible ECPair abstraction that provides .sign etc as available now, and then users who don't use that can simply do whatever they want, provided they return a 64-byte R|S encoded signature.

[dcousens]

Closing in favour of #1048, #1045 and #1049 (and #1046)