Truncate long tokens in default_validator's judge message. #346

[gkreitz]

This updates default_validator to avoid creating huge judgemessage.txt when the token containing a difference is huge (e.g., the user forgot to output spaces).

When printing single tokens, we cut after 30 bytes, and when printing two tokens, we print up to 15 bytes of the common prefix ... first 15 bytes from where the strings first differ.

While we can't assume anything about the output encoding, we make a small attempt not to insert our "..." in the middle of a utf-8 character, as that is by far the most commonly used encoding. (You can still get weird efffects here if using composition characters, as I don't want to add a dependency on a full unicode handling library).

Also adds test cases to cover this truncation behavior, plus some more test cases, so we now at least should have one test case per different judgemessage we currently canemit.

Fixes #346