Describe the bug
Normal and Laplace accept non-positive values for their required scale parameters and can return mathematically invalid pdf values instead of rejecting the input.

For example:

• Normal(mu=0, sigma=-1).pdf(0) returns -0.3989422804014327

• Laplace(mu=0, scale=-1).pdf(0) returns -0.5

A probability density function should never be negative, so this is a silent correctness issue.

To Reproduce

from skpro.distributions.normal import Normal
from skpro.distributions.laplace import Laplace

print(Normal(mu=0, sigma=-1).pdf(0))
print(Laplace(mu=0, scale=-1).pdf(0))

Observed output in skpro:

-0.3989422804014327
-0.5

SciPy comparison:

from scipy.stats import norm, laplace

print(norm.pdf(0, loc=0, scale=-1))
print(laplace.pdf(0, loc=0, scale=-1))

SciPy returns:

nan
nan

Expected behavior
These distributions document sigma and scale as positive parameters. Passing non-positive values should raise a clear ValueError, ideally at construction time, instead of allowing invalid parameters through to pdf evaluation.

Environment
OS: Windows
Python: 3.9.13
skpro: current main branch (local clone)

Additional context
This appears to be missing parameter validation in the distribution constructors. I checked a few related distributions as well; the strongest confirmed cases from direct reproduction are Normal and Laplace.

I'd be happy to open a PR for this.