cipher: fix seeking implementation in the stream cipher wrapper

[newpavlov]

The change forbids seeking to the last keystream block and its application. Additionally, unchecked methods added as an escape hatch.

Closes #1808

[tarcieri]

Is this really a fix for #1808? I would think a real fix would make the last keystream block usable

[newpavlov]

See my comment in the issue. I spent a significant time trying to implement the last byte approach, but I couldn't get a satisfying solution.

[newpavlov]

Another alternative could be to add a special "exhausted" state to the wrapper (either as a separate field, or as part of ReadBuffer), but I don't quite like it either. Adding the field would make all buffered stream cipher implementations bigger (often by more than one byte because of alignment), while modifying ReadBuffer would mean that we need to check the special case every time. Also, I would like to keep the block-buffer code as simple as possible.

Considering that users arguably should not rely on the last keystream block in practice, I am not sure it's worth the trouble.

[newpavlov]

@tarcieri
Could you take a look at the updated docs?

[tarcieri]

Docs look good to me