Restore sorted cell iteration order with lazy cache

[ken-swyfft]

Summary

• After Replace SortedDictionary with Dictionary in HSSFRow, XSSFRow and SXSSFRow #1753 replaced SortedDictionary with Dictionary in XSSFRow, GetEnumerator() and Cells no longer returned cells in ascending column order — a breaking change for existing consumers who foreach over rows
• Add a lazily-built sorted cell cache (List<ICell>) that is populated on first read and invalidated on mutation (add/remove/rebuild)
• Preserves the Dictionary O(1) insert/lookup performance win while restoring the sorted iteration contract
• Applied consistently to XSSFRow, HSSFRow, and SXSSFRow

Context

Raised by @tonyqus in #1742 (comment) — the SortedDictionary → Dictionary change in #1753 means cells returned from GetEnumerator are no longer sorted by column index, which could break existing NPOI-based logic that assumes sorted iteration.

Design

• Reads dominate writes (sheets are built once, iterated many times), so the sort is cached and reused across multiple iterations
• Cache is invalidated (null-ed) on any cell mutation — CreateCell, RemoveCell, RebuildCells
• In-flight enumerators see a stable snapshot (the old list), which is safer than the old SortedDictionary behavior that would throw InvalidOperationException on concurrent modification
• XSSFRow.OnDocumentWrite() also reuses the cache instead of re-sorting

Test plan

• TestCellIterationOrderWithSparseColumns passes (net8.0 + net472)
• All XSSFRow tests pass: 34/34
• All HSSFRow tests pass: 13/13
• All SXSSFRow tests pass: 16/16

🤖 Generated with Claude Code

[tonyqus]

LGTM