Add Float.round function

[bitwalker]

Float.round allows rounding floats to arbitrary precision, as well
as specifying how midpoint rounding is performed (up/down).

This should provide accurate rounding for up to 15 digits of
precision (the max guaranteed by the IEEE-754 standard for
64-bit floating point numbers). If you can think of anything my
tests won't catch, let me know.

[bitwalker]

I initially wanted to place this next to the Kernel.round function, since they do the same thing (more or less), and it might've been useful to have in guard clauses for the same reasons Kernel.round would be, but my initial effort there failed when I couldn't use and, in, and presumably .. at the point where Kernel.round is defined. Anyways, Float seems like a good module to put this in, and might make more sense anyways.

[josevalim]

Please mention in the docs that precision needs to be in between 0 and 15. :)

[josevalim]

I agree this belongs to float. Thanks for the pull request! :D

[bitwalker]

Added the clarification on the available range for precision. Let me know if you spot anything else!

[bitwalker]

Thanks!