[BUG] Pareto _pdf and _log_pdf return nonzero values outside support (x < scale)

[ANANYA542]

Describe the bug
The Pareto distribution is only defined for x >= scale. For x < scale, the PDF should be exactly 0. However, Pareto._pdf and Pareto._log_pdf in skpro/distributions/pareto.py blindly evaluate the PDF formula for all x, returning large wrong positive values for inputs outside the support.
For example, with alpha=3, scale=2, calling _pdf(1.0) returns 24.0 instead of 0.0.
The _cdf method in the same file already correctly handles this boundary using np.where(x < scale, 0, ...), so the fix pattern is straightforward.

To Reproduce

import numpy as np
from scipy.stats import pareto as scipy_pareto
from skpro.distributions.pareto import Pareto
alpha, scale = 3.0, 2.0
d = Pareto(alpha=alpha, scale=scale)
test_x = [0.5, 1.0, 1.5, 2.0, 3.0, 5.0]
print(f"{'x':>5}  {'in support?':>12}  {'skpro _pdf':>12}  {'scipy pdf':>12}")
for x in test_x:
    x_arr = np.array([[x]])
    skpro_val = float(np.asarray(d._pdf(x_arr)).flat[0])
    scipy_val = scipy_pareto.pdf(x, alpha, scale=scale)
    in_supp = "YES" if x >= scale else "NO"
    print(f"{x:>5.1f}  {in_supp:>12}  {skpro_val:>12.4f}  {scipy_val:>12.4f}")

Output:

    x   in support?     skpro _pdf     scipy pdf
  0.5            NO     384.0000        0.0000
  1.0            NO      24.0000        0.0000
  1.5            NO       4.7407        0.0000
  2.0           YES       1.5000        1.5000
  3.0           YES       0.2963        0.2963
  5.0           YES       0.0384        0.0384

_log_pdf has the same issue — returns finite positive values instead of -inf:

  x=0.5: skpro=5.9506  scipy=-inf
  x=1.0: skpro=3.1781  scipy=-inf

Screenshot of full local reproduction attached below:

Expected behavior
1._pdf(x) should return 0.0 for any x < scale, matching scipy.stats.pareto.
2_log_pdf(x) should return -inf for x < scale.

Environment

• OS: macOS
• Python: 3.11
• skpro: latest main branch

Additional context
The _cdf method in the same file (line 145) already handles the support boundary correctly:

cdf_arr = np.where(x < scale, 0, 1 - np.power(scale / x, alpha))

The _pdf and _log_pdf methods simply need the same np.where guard added — a one-line addition per method.

[ANANYA542]

@fkiraly — noticed that Pareto._pdf returns nonzero values for x below the scale parameter. For example pdf(1.0) = 24.0 when scale=2.0, should be 0.0. The _cdf method in the same file already handles this correctly with np.where — _pdf and _log_pdf just need the same guard. After the fix, pdf(1.0) correctly returns 0.0 and log_pdf(1.0) returns -inf as expected. Will be opening a PR showing the before/after outputs if this looks good to you!