Replace obsolete isinff(...)

[fxing-qnx]

Fixes # N/A

Proposed changes: Replace isinff(...) with a combination of isinf(...) and signbit(...)

Benchmark results: No changes in performance.

Test suite changes: N/A

Almost the same to #188, but more aggressive. isinff(...) is completely remove from the source tree.
Reason: Mentioned in PR188, isinff(...) is both obsolete and non-standard. To maximize portability and suggested by GNU, we should have it replaced.

PS, Sorry due to our process we need to always fork under tags even if a tag only contains version changes. I have a rebased branch ready for merge now.

Edit:
Fix #278

[ebassi]

4/23 graphene / box ERROR 0.02s exit status 1

▶ 4/23 - box(empty).center() to be in (0, 0, 0) FAIL

[fxing-qnx]

4/23 graphene / box ERROR 0.02s exit status 1

▶ 4/23 - box(empty).center() to be in (0, 0, 0) FAIL

Yeah I saw that, very weird.........

[fxing-qnx]

@ebassi -ffast-math is set by build configurations. According to https://gcc.gnu.org/wiki/FloatingPointMath, this flag will also set -ffnite-math-only which will violate IEEE's floating-point specification and can potentially cause standard functions like isinf(...) to fail.

The problem on the surface is isinf(...) will only work for double type instead of float. For some reasons isinff(...) will return the right answer even with those flags.

[ebassi]

The MSYS2 build failure in CI is caused by a package rename, and has been fixed in master.

[fxing-qnx]

It should now be fine........

[fxing-qnx]

@ebassi Are the changes ok now?