Consolidate apfloat "round with ties to nearest even" logic

During code review of https://github.com/google/xls/commit/f8853882164c629c39aa3fb75336d4d7615d1729, we discussed that there would now be multiple (different) implementations of "should we round up (with ties to nearest even)" for apfloats. Specifically, we have:

https://github.com/google/xls/blob/53f30d068f9bb9e7888ecbe4c34273f5c3223e27/xls/dslx/stdlib/apfloat.x#L280-L298

https://github.com/google/xls/blob/53f30d068f9bb9e7888ecbe4c34273f5c3223e27/xls/dslx/stdlib/apfloat.x#L794-L806

We left it as is because of a few considerations:

1. The implementations are different, and it would be nice to compare QoR.
2. Neither implementation is part of the public API.
3. `rne` (unintentionally?) adds behavior for `lsb_idx` overflowing the input fraction size (https://github.com/google/xls/blob/53f30d068f9bb9e7888ecbe4c34273f5c3223e27/xls/dslx/stdlib/apfloat.x#L311), while the latter knows this will never happen by construction.
4. `rne` may benefit from renaming since it doesn't do any "rounding" as implied by the name.