Fix #1140: Align the partial hierarchy modes on row labels in the pandas engine

[albertohernandez1995]

Summary

With the pandas engine, hierarchy in partial_null / partial_zero mode crashed when a rule's right-side code items were missing for some values of the other identifiers:

• HAAssignment used the operands' row labels as positions (iloc), which fails once the partial mask drops rows that are not the last ones. It now selects the rows by label.
• update_partial_data widened the mask with |=, keeping only the index of the first code item. A code item computed by an earlier rule may miss some groups, so the mask now covers the union of both indexes, and visit_HRBinOp filters each operand by label.

The pandas results now match the DuckDB engine. Added tests GH_1140_1–GH_1140_4 (both cases from the issue, in both partial modes).

Checklist

• Code quality checks pass (ruff format, ruff check, mypy): mypy's 3 errors in Model/__init__.py are already on 1.9.X, not from this change
• Tests pass (pytest)
• Documentation updated (if applicable)

Impact / Risk

• Breaking changes? No. Only the pandas engine changes, and only in cases that used to raise an internal error.
• Data/SDMX compatibility concerns? None.
• Notes for release/changelog? pandas engine: hierarchy in partial_null and partial_zero modes no longer crashes when code items are missing for some identifier values.

Notes

The DuckDB engine is unchanged.

---

Closes #1140

[javihern98]

Looks good, thanks! 😊