fix sin/cos and (/) behavior

[massma]

This hopefully resolves #59 by constraining the infimum in cos to [0, pi).

It also changes (/) to improve behavior with zero end points (see #19 ). Before:

> 1.0 / (0.0 ... 1.0)
Infinity ... Infinity
> 1.0 / (-1.0 ... 0.0)
-Infinity ... Infinity

After:

> 1.0 / (0.0 ... 1.0)
1.0 ... Infinity
> 1.0 / (-1.0 ... 0.0)
-Infinity ...-1.0

For what it's worth, the new behavior is consistent with the Multiple Precision Floating-Point Interval (MPFI) library. Because there might be opinions, I didn't touch 1 / singleton 0.0 (e.g., #47), but MPFI does return whole for that, rather than an exception.

Also do you have any interest in defining recip in terms of (/)? I thought there might be a reason recip is defined as is, but if not, redefining recip would fix #53.

Finally, I am a novice to Haskell so all of this might benefit from a second set of eyes.

Thanks for the help and putting together this package!

Adam

EDIT: I am leaving this pull request open for now, but I also have a branch going:

https://github.com/massma/intervals/tree/adam-fork

of some (possibly opinionated and novice) changes I made to address the bugs I needed to for my application. So far this branch fixes #19, #47, #53, #59, #61. If you want any of these updates to be incorporated into intervals just let me know: I am happy to filter and edit the branch according to your preferences, and submit a fresh pull request!