Clarify the halfwidth mean-error difference sign

[Kkkakania]

Description

Clarifies the sign convention behind the difference returned by halfwidth_ci_test.

The function consistently returns score(forecasts2) - score(forecasts1), but each score follows the selected metric argument convention. For mean error, xskillscore defines me(a, b) as mean(a - b) and halfwidth_ci_test calls it as me(observations, forecast). The resulting sign is therefore opposite to calculations that define error as forecast - observation. The sign does not change the significance decision because that comparison uses the absolute difference.

This documents the behavior reported in #413 without changing the existing statistical calculation.

Closes #413

Type of change

• Bug fix (documentation clarification; no runtime behavior changes)

How Has This Been Tested?

• uv run --extra test pytest --doctest-modules xskillscore/core/comparative.py -q (2 passed)
• pre-commit on xskillscore/core/comparative.py (all applicable hooks passed)

Checklist (while developing)

• The updated function documentation builds as a valid doctest module.

Pre-Merge Checklist (final steps)

• Based on the current upstream main.