MAINT: special: add an explicit NaN check for iv arguments

[person142]

Reference issue

Closes gh-10656.

What does this implement/fix?

This adds a check for NaN arguments in special.iv and returns NaN in those cases.

Additional information

@RizzerOnGitHub can you confirm that this fixes your issue?

[drew-parsons]

Looks good. test_nan_inputs[iv] now passes on all architectures,
see https://buildd.debian.org/status/package.php?p=python-scipy&suite=experimental

(there are various failures in other tests, which for some reason don't cause the Debian builds to fail for hppa, riscv64, sparc64. But test_nan_inputs[iv] now passes)

[pv]

@RizzerOnGitHub: if there are now failures for other functions in test_nan_inputs, could you update and reopen the issue gh-10656, thanks!

[drew-parsons]

Thanks @pv . Just to clarify, all test_nan_inputs tests now pass. The other failing tests are in other tests or other submodules.

[drew-parsons]

And looking more closely at the other test failures, most of them seem to be a question of tolerance, akin to the problems reported in Issue #7535

[person142]

If it's a small tolerance issue we usually just bump the tolerances a bit since floating point determinacy is not always as good as one would like.