[ENH] Sync skpro and sktime probabilistic metrics modules: sample_weight and output consistency

[Omswastik-11]

Summary

This PR brings the skpro probabilistic metrics module into alignment with sktime, as discussed in issue #367.
It adds sample_weight support across metrics, ensures consistent output types, and fixes weighted multioutput behavior.

---

Changes

1. Sample Weight Support

• Added a sample_weight parameter to:

  • BaseProbaMetric.evaluate
  • BaseDistrMetric.evaluate

• Updated input validation methods:

  • _check_consistent_input
  • _check_ys
    to correctly propagate sample_weight

• Implemented proper weighted averaging using np.average in:

  • _evaluate (Proba metrics)
  • evaluate (Distribution metrics)

---

2. Output Consistency Improvements

• BaseDistrMetric.evaluate now returns a pd.Series when:

  • multioutput="raw_values"
  • the metric is univariate (non-multivariate)

• This matches:

  • BaseProbaMetric behavior
  • sktime conventions

• Fixes the previous inconsistency where a 1-row DataFrame was returned.

---

3. Weighted Multioutput Fixes

• Improved handling of array-like multioutput weights in BaseDistrMetric.evaluate
• Ensures correct weighted aggregation across variables
• Fixes incorrect averaging when users pass custom weights.

---

Tests Added / Updated

• test_sample_weight_pinball (in test_probabilistic_metrics.py)
  Verifies sample_weight support for quantile and interval metrics.

• test_sample_weight_logloss (in test_distr_metrics.py)
  Verifies sample_weight support for distribution-based metrics.

• test_multioutput_weights_logloss (in test_distr_metrics.py)
  Ensures weighted multioutput aggregation behaves correctly.

• Updated test_distr_evaluate to assert that raw-value outputs are returned as pd.Series.

Fixes #367

[fkiraly]

Sorry if the description of the task has not been precise enough, I think you did not quite catch what should happen in the TestAll classes.

The TestAll classes run their tests on all the metrics, and should check API elements that are common to all. So, if you add sample_weight as an argument, the TestAll classes should pass examples where sample_weight is being passed.