get_xlsx_sheet_info's unreadable-workbook path and the malformed @ENCODING skip are untested #1006

Where: faircode/loaders_extra.py (get_xlsx_sheet_info() line 144, held_out_encoding_ignored_notes() lines 169-170), tests/test_loaders.py.

Current gap: get_xlsx_sheet_info() is meant to return None when a workbook cannot be opened or has no sheets, so the CLI and MCP notice stays silent and the real read error is reported instead; and held_out_encoding_ignored_notes() must skip a spec with a malformed @ENCODING suffix because the real parse reports it. Coverage shows neither branch runs, and a refactor that let these exceptions escape would turn a clean error into a traceback.

Repro:

$ python3 -m pytest tests -q -m "not slow" --cov=faircode --cov-report=term-missing | grep loaders_extra.py
faircode/loaders_extra.py ... 144, 169-170

Suggested approach: a corrupt .xlsx (a text file with that suffix) asserting None, and held_out_encoding_ignored_notes(["h.xlsx=race@"], "latin-1") asserting [] and no exception.