* Fixes a bug in shikata where input greater than 0xffff length would
still use 16-bit counter
* Short circuits finding bad xor keys if there are no bad characters to
avoid
* Fixes huge performance issue with large inputs to xor-based encoders
due to the use of String#+ instead of String#<< in a loop. It now
takes ~3 seconds on modern hardware to encode a 750kB buffer with
shikata where it used to take more than 10 minutes. The decoding side
takes a similar amount of time and will increase the wait between
sending the second stage and opening a usable session by several
seconds.
I believe this addresses the intent of pull request 905
[See #905]
b3b68c1b90