Fix HMAC usage

[dd86k]

While the Digest API uses std.digest.hmac.HMAC for HMAC, which adds a secret hashed key using the template API, the way BLAKE2 does is just... Weird.

At initiation:

• Parameter block index 0 byte 1 (little-endian) sets key size in bytes.
• Key is XOR'd with state alongside IV (ulong for b, uint for s, see the inner_t alias).

Which is simply not possible with the HMAC structure template. Unless there's a hack I don't know of.

[dd86k]

I think I could do a void key(scope const(ubyte)[] data...) function that performs a "second stage" initiation.

[dd86k]

Got key stuff in but that's not hmac so keeping this open.

[dd86k]

For the HMAC templates to work, the key needs to be initiated at compile-time. The issue is, while the default key is the IV, they are of different types in the s (uint) and b (ulong) variants.