Improve performance of `cycles` and `sign`

[dkarrasch]

I came across this package in search of a quick sign implementation. I noticed a few optimization possibilities in cycles. Moreover, I found the sign implementation very inefficient, because all you care about is the length of the result of cycles(p), so one can take the same code and replace every array concatenation by a += 1. That reduces memory allocation and yields significant performance improvements.

[scheinerman]

Thanks @dkarrasch . I've merged your changes and will release as a new version.