BUG: Fix stats.skewnorm.ppf for large skew values

[maresb]

Reference issue

Closes gh-20124

What does this implement/fix?

Fix an upstream bug in Boost that results in nonsensical results when computing PPF of skewnorm distributions that have large values of skew.

I'm trying to verify my regression test on my fork https://github.com/maresb/scipy/pull/3.

[lucascolley]

x-ref gh-20122

[maresb]

Ah, it's a bit disappointing that upgrading Boost didn't automatically solve gh-20124. But at least the warning message is different.

Now:

RuntimeWarning: Error in function boost::math::quantile(const skew_normal_distribution<d>&, %1%): Unable to locate solution in a reasonable time: either there is no answer to quantile or the answer is infinite.  Current best guess is %1%

Previously:

RuntimeWarning: overflow encountered in _skewnorm_ppf

I may have to chase this one further upstream in Boost.

[maresb]

Superseded by #20643, thanks @dschmitz89!!!