Fix errors after numerics reform

[ruuda]

There are still lots of warnings about Zero and One being deprecated, but I do not see an easy way of resolving those without splitting everything in a Float and an Int case.

[mbrubeck]

I think we should just change callers of Zero and One to use things like Length(0.0). Are there any cases where they need to be generic?

[pcwalton]

@mbrubeck The matrix identity function uses those, for example.

[ruuda]

What about introducing a Ring trait that is Add and Mul and has one() and zero()?