Add cdf/ccdf/quantile to SkewNormal via Owen's T

[devmotion]

Stacked on #2070. This PR is based on #2070 (Roots-based quantile root-finding) and
targets dmw/roots; its diff shows only the SkewNormal feature. The near-zero-root
convergence that SkewNormal needs (its support straddles 0) is provided by #2070's solver,
so this PR no longer touches src/quantilealgs.jl. Merge #2070 first.

Summary

SkewNormal was missing cdf/ccdf (the source had #cdf requires Owen's T function.). StatsFuns 2.2 now provides owens_t, which unblocks the closed form:

cdf(x)  = Φ(z) - 2·T(z, α)
ccdf(x) = Φ̄(z) + 2·T(z, α),   z = (x - ξ)/ω

This PR adds cdf/ccdf, refactors pdf/logpdf for consistency, and adds quantile/cquantile. logcdf/logccdf/invlogcdf/invlogccdf use the generic fallbacks.

• Requires StatsFuns 2.2; bumps the package version.

Quantile root-finding (handled by #2070)

quantile/cquantile are Newton iterations from the mode via quantile_newton/cquantile_newton. The original version of this PR also patched src/quantilealgs.jl with an absolute-tolerance floor, because the hand-rolled relative-only stopping rule never converges as the root approaches zero (which SkewNormal hits). That fix is now subsumed by #2070, which replaces the hand-rolled loops with Roots.find_zero and so converges for near-zero roots out of the box — this PR therefore drops its own quantilealgs.jl change.

invlogcdf/invlogccdf are intentionally left to the generic quantile(d, exp(lp)) fallback rather than the @quantile_newton macro: the log-space invlogcdf_newton overshoots into the underflowed tail from the approximate mode m_0 and returns NaN for extreme log-probabilities (e.g. p ≲ 1e-8), even with #2070's bracketing fallback, whereas routing through quantile_newton stays robust there.

Tests

• Added a sn-based reference class test/ref/continuous/skewnormal.R (moments from sn:::sn.cumulants, an independent computation rather than a reimplementation of the Julia formulas), 4 entries to continuous_test.lst, the generated entries to continuous_test.ref.json, and pinned sn + deps in renv.lock.
• The dedicated testset is reduced to construction, support boundaries, and the α=0 reduction to Normal.

Verification

• R reference framework: 88332/88332 pass for the SkewNormal entries.
• Reference values cross-checked: R sn and Julia agree to ~1e-14.
• quantile/cquantile round-trips and the invlogcdf tail fallback verified against Simplify quantile root-finding to Roots defaults #2070's solver (incl. roots straddling 0).

🤖 Generated with Claude Code

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 86.89%. Comparing base (6b5aa0d) to head (9632e69).

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #2069      +/-   ##
==========================================
+ Coverage   86.74%   86.89%   +0.14%     
==========================================
  Files         149      149              
  Lines        8882     8898      +16     
==========================================
+ Hits         7705     7732      +27     
+ Misses       1177     1166      -11     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.