Avoid spurious overflow and divide by zero in ibeta.

[jzmaddock]

Add tests extracted from SciPy bug report.
Fixes #1006.
See also #1009.

[mdhaber]

Thanks @jzmaddock! This seems to help on my M1 mac.

import scipy.stats
scipy.stats.beta(a=1e-323, b=5).ppf(.2)

Before this patch, I get an error; after this patch, I get 1.8189894035458565e-12. I'm not sure if the relative accuracy is great, since taking the cdf at that argument returns 1.0 instead of 0.2, but I'm guessing it will fix the crash I experience on other machines.