fix: handle pandas extension dtypes in ownership calculations

[monody0007]

Pandas 3 text columns and pandas nullable numeric dtypes can make NumPy's dtype check raise TypeError in ownership calculations. Use pandas' dtype predicate at both affected sites while retaining the numeric-text fallback and nonnumeric results.

Fixes #1368. Thanks to @manantlerio for the reproduction, affected-site analysis, and proposed fix. The empty-trades change in #1367 remains separate.

The regression exercises Form4.parse_xml() with the checked-in Snowflake fixture modified only in memory. It covers footnoted text, unreadable shares, nullable numeric columns, and the normal share total; the modified XML is synthetic and no cassette is recorded.

Validation:

• New regression: 11 passed on pandas 3.0.6 and pandas 2.3.3. On the original code: 7 failed/4 passed on pandas 3; 5 failed/6 passed on pandas 2, all at the dtype check.
• pytest -p tests._offline_harness -p no:pytest-retry -q tests/issues/regression/test_issue_1368_ownership_numeric_dtypes.py
• Offline ownership fast tests: 9 passed, 50 deselected. Changed-file Ruff and regression provenance/skip/changelog checks passed.
• Existing whole-repo Ruff and production-file Pyright findings match the base. Live SEC and the full network suite were not run.

AI assistance: coding, test execution, and an independent model review were performed with AI tools. This is disclosed as AI-assisted work; no human code review is claimed.

[monody0007]

Merged current main (73491ae) into this branch to resolve the conflict with #1367. The resolution retains its None/empty guard in shares_traded and the pandas-aware dtype check from this PR.

The #1368 and #1359 regression files pass on both pandas 3.0.6 and 2.3.3: 15 tests per environment, with no skips. The wider network-dependent tests remain unverified locally; their offline failures match the unchanged upstream base.