Avoid using floor in rem_pio2

[quaternic]

Possible workaround for #976 (comment)

I'm not familiar with the details of this (rather complex) algorithm and implementation, but the comments claim that x must be positive, and that would seem to imply that the z in floor(z) also is.

[folkertdev]

This passes the test suite (just needs a formatting fix), and with this change the floor/ceil assembly versions also pass the test suite as expected.