Isolate worksheet-local property mappings to prevent cross-sheet metadata corruption and spurious header ingestion

[Montana]

Hi Homeland Security,

This PR fixes a worksheet-boundary state management defect in TemplateReader.inventory_records() that can corrupt imported metadata or terminate Excel ingestion with a KeyError.

The property mapping is currently initialized outside the worksheet loop. After processing the first inventory worksheet, subsequent worksheets inherit its column-to-property associations and bypass header initialization.

This produces three observable failures:

• Subsequent worksheet headers are imported as inventory records.
• Reordered columns silently associate values with incorrect DCAT properties, including assigning dataset titles to identifiers.
• Additional populated columns absent from the original mapping raise a KeyError.

The change moves props = None inside the worksheet loop. Each inventory worksheet independently initializes its mapping from its own header comments and skips that header before processing records. Existing record construction and blank-row handling remain unchanged.

The accompanying parametrized regression tests exercise eight combinations: identical, reordered, expanded, and reduced column layouts, each with either a populated or header-only first inventory worksheet. The fixtures also include blank rows and a non-inventory instructions sheet.

Cheers,
Michael