Allow scalar distributions to accept vector inputs

[Aymuos22]

Summary

• Allow scalar (0‑D) distributions to accept vector inputs by relaxing the coercion helpers so they no longer broadcast every input to shape (), while still returning scalars for scalar calls.
• Adjust the boilerplate to only unwrap true 0‑D numpy outputs, ensuring vectorized results remain arrays.
• Add regression tests for scalar Normal and Poisson showing that pdf/log-pdf/cdf (and pmf/log-pmf) now work on numpy arrays without manual loops.

Testing

• pytest skpro/distributions/tests/test_base_scalar.py