Use faster AES_CTR, SHA1 implementations?

If we can make these go any faster than they do now, we'll see it in our profiles; they are two of the biggest chunks of our usage.

They're already optimized and using assembly, but the numbers below show that we could possibly do better.