nonce byte array allocations

[nbrownus]

nonce preparation in Encrypt and Decrypt causes an allocation. In an application encrypting/decrypting at 200k+ times a second this becomes rather noticeable in the garbage collector.

One way I've been thinking of working around this is to provide a function similar to this:

func (c aeadCipher) EncryptNB(out []byte, n uint64, ad, plaintext []byte, nb []byte) []byte {
	nb[0] = 0
	nb[1] = 0
	nb[2] = 0
	nb[3] = 0
	binary.BigEndian.PutUint64(nb[4:], n)
	return c.Seal(out, nb, plaintext, ad)
}

Obviously there are a handful of issues here, is the byte array big enough, too big, being used by another thread? Adding (some of) those checks would still likely be less overhead though.

I can bring crypto into my part of the program, but this feels bad:

nb[0] = 0
nb[1] = 0
nb[2] = 0
nb[3] = 0
binary.BigEndian.PutUint64(nb[4:], n)
return s.c.(cipher.AEAD).Open(out, nb, ciphertext, ad)

Other thoughts would be to internalize a heap for storing nonce byte arrays.

Wanted to get your thoughts on the matter before I sent any PRs your way, thanks!