Remove `vec_arith()` numeric `y` methods

[DavisVaughan]

I'm thinking about how to add custom double dispatch for vec_arith(), and I think the first step is to try and clean up existing numeric methods by separating them into integer/double.

We can't touch any methods that are of the form vec_arith.numeric.*() yet, because we don't export vec_arith.integer or vec_arith.double (and I don't think we want to).

However, we can replace vec_arith.*.numeric() methods with their explicit integer/double counterparts. That is what this PR does.

It also leaves in some FIXME commented out sections that will be removed in a subsequent PR when custom double dispatch is added.

Tests have been updated to be more explicit about testing both integer and double cases.

[lionel-]

Probably best to leave this for 0.4.0.

[DavisVaughan]

Yea definitely

[DavisVaughan]

This is probably the right first step, but we will rebuild it from scratch when we tackle https://github.com/r-lib/vctrs/issues/1063 for real