Thanks for the review.
Yes. As Johannes pointed out, the word_regex was very loose and did not
properly tokenise the code.

I have fixed(and added tests) for the following things in this patch:

+= or -= is not split into + = or - =
-a or +a is split into 2 tokens(for each)

Since kotlin supports `_` in between digits to improve readability,
therefore numbers like `100_000` are single tokens and not broken up
by the regex in this patch.

Kotlin also does not support octal literals, and so the regex does not
attempt to find it.