Add Excel-compatible nested SUBTOTAL handling

[tomtomau]

Summary

Makes SUBTOTAL ignore cells within its reference range that themselves contain a SUBTOTAL formula, preventing double-counting of nested subtotals — matching Excel and Google Sheets behaviour.

Previously SUBTOTAL dispatched to the same aggregation helpers as SUM/AVERAGE/etc. and treated nested SUBTOTAL cells like any other value, so an outer subtotal spanning sub-group subtotals double-counted them.

Approach

• A skipSubtotalCells flag is threaded through the shared aggregation chain (do* → reduce/reduceAggregate → evaluateRange → getRangeValues), defaulting to false so SUM/AVERAGE/COUNT/etc. are unchanged.
• getRangeValues() and the single-cell-reference path skip cells where a new isSubtotalCell() helper detects a top-level SUBTOTAL call in the cell's formula AST (canonical name comparison, so i18n-safe).
• Subtotal aggregation uses a dedicated range-cache key (SUBTOTAL_<fn>) so it never collides with a plain aggregation over the same range.

Applies to both the 1–11 and 101–111 code families. Hidden-row handling (the practical difference between those families) remains out of scope, as the engine is headless and has no row-visibility concept.

Tests

New spec test/unit/interpreter/function-subtotal.spec.ts covering: baseline parity with SUM, nested exclusion for sum/average/count, the 1xx code family, SUM-vs-SUBTOTAL cache independence, single-cell references, and recalculation on edit.

• npx jest function-subtotal → 8/8 pass
• smoke suite → pass; npm run lint → 0 errors; npm run compile → clean