Remove the rem_float intrinsics

[gbaraldi]

Since #47501 we have native versions of the rem functions, but the original PR didn't change the fastmath version. It turns out that OpenLibm's fastmath version was already slower than the non fast one, so add a fast version of it and remove the intrinsics.

[gbaraldi]

@nanosoldier runtests()

[ViralBShah]

The win64 test had some failures and then eventually timed out - but the failure is not being detected right in the CI.

https://buildkite.com/julialang/julia-master/builds/21178#01865601-bb06-4472-910d-f58e0dc9c443/904-1109

cc @DilumAluthge

[DilumAluthge]

Hmmm. In this case, it seems like the timeout worked correctly, right? The win64 tests timed out at 3 hours, so Buildkite killed the job and exited with status code 1. Then Gabriel retried the failed job; the retry succeeded.

[DilumAluthge]

However, on Viral's other PR, Windows timed out, and Buildkite exited with status code zero, which is obviously a big problem.

@staticfloat has sent a support ticket into Buildkite; we're waiting to hear back.

[ViralBShah]

Oh I didn't realize that. I see the successful job next.

[nanosoldier]

Your package evaluation job has completed - possible new issues were detected.
A full report can be found here.

[gbaraldi]

It seems the only package that used the intrinsic is Unitful. They seem to use the intrinsics directly which seems a bit odd.

[ViralBShah]

I suppose that since the Unitful.jl PR is merged, this PR should be good to merge.