[BUG] fix QPD_U pdf scaling and symmetric quantile triplet case

[VividhDesign]

Reference Issues/PRs

Fixes #1187

What does this implement/fix? Explain your changes.

Two commits.

1. _pdf multiplies by width, which was missing from the chain rule.
2. For a symmetric SPT, sinh / arcsinh in _ppf, _cdf and _pdf are replaced by their limit, the identity.
   • With the existing delta = 1, kappa = H - B, gamma = 0, xi = B, this gives ppf(p) = B + (H - B) * phi.ppf(p) / phi.ppf(1 - alpha).
   • That recovers the three quantiles and matches the limit of nearly symmetric triplets.
   • _prep_qpd_vars now also returns a symmetric mask for mode U, so this works entry-wise for array parameters.

Does your contribution introduce a new dependency? If yes, which one?

No.

What should a reviewer concentrate their feedback on?

• Whether the limit is the behaviour you want for the symmetric case. An alternative is kappa = (H - B) / sinh(1): it matches the three quantiles but isn't continuous with the nearly symmetric case.

Did you add any tests for the change?

Yes, in test_qpd.py:

• test_qpd_u_pdf_is_cdf_derivative: pdf vs. numerical derivative of cdf, and the integral equals 1.
• test_qpd_u_symmetric: recovers the quantiles, matches the normal limit, and is continuous with the nearly symmetric case.

Both fail on main. skpro/distributions passes (8586).

Any other comments?

(AI-assisted; reviewed and tested by me.)

PR checklist

• The PR title starts with [BUG]