Replace CSHA256::FinalizeNoPadding with CSHA256::Compress

[daira]

In #814 we fixed a potential usage error of the SHA-256 hash API when used to obtain the compression function, by requiring 64 bytes to have been written (if a particular flag is passed in). However the resulting construct/Write/FinalizeNoPadding API is unnecessarily complex and verbose.

Replace it with a simple function:

static void CSHA256::Compress(unsigned char (&hash)[OUTPUT_SIZE], const unsigned char (&input)[BLOCK_SIZE]);

(Not 100% sure I have the reference syntax correct there; the intent is a reference to an array of known size, which is a const array in the case of input.)

[defuse]

I suggested this to @ebfull in our pairing and he said he wanted to be able to use .Write() with the compression function (e.g. to write the a_sk first and the other stuff next).

I actually think we should do this anyway, because then we aren't relying on .Write() and the complicated check (bytes == 64) to make sure the compression function is actually getting called.

[daira]

I considered that, but I think it is mixing up the concatenation of byte strings with the application of the hash. I would much prefer that we had a concise (and more functional in style) way to do the former, independent of the latter. We do concatenation in many more cases than we do hashing.