fix gcm chunked

[Graveflo]

#92

[Graveflo]

Adding chunk support for the ghash as well so the tag matches. This is still 0 copy for one-shot calls (except for the very last 16 byte increment if misaligned) and for well-behaved 16 byte aligned chunked input.

[Graveflo]

also swapping the order of hashing and cipher run to support in-place encryption and decryption. The hash gets corrupted otherwise.

[cheatfate]

Thank you very much for you work, but i found another way in #94