Add (*File).Recalc to evaluate formulas and persist cached values

[krystophny]

Problem

excelize has no "open → recalc → save" workflow. A consumer that wants it hits three walls:

1. No cached-value writer. SetCellFloat / SetCellStr / etc. call removeFormula, which deletes <f> and wipes shared-formula children.
2. Round-tripping. CalcCellValue returns a string; writing it back via SetCellValue records t="s". Aggregates over that shared-string blob return 0.
3. 3D references:
   • SUM(Sheet1:Sheet3!A1) unquoted → #NAME? invalid reference.
   • SUM('Sheet1':'Sheet3'!A1) quoted → parses, but rangeResolver overwrites sheet per iteration and evaluates only the last sheet.

Proposal

Add two exported methods:

func (f *File) RecalcCell(sheet, cell string) error
func (f *File) Recalc(opts ...RecalcOptions) error

RecalcCell evaluates one formula cell and writes the typed result into <v>/<t>, preserving <f> and shared-formula grouping. Recalc walks formula cells and calls RecalcCell on each. Dependency resolution reuses the existing recursive calcCellValue evaluator; circular references are bounded by MaxCalcIterations.

Scope

Four narrow PRs split under this umbrella, each with its own tracking issue where appropriate:

1. 3D references, unquoted form. parseReference recognises <sheetA>:<sheetB>!<ref> and expands in workbook order; aggregates consume the expanded matrix. Covers SUM(Jänner:Dezember!M40) etc. Closes this issue (partial).
2. RecalcCell + private setCellCachedValue helper + ErrCellNoFormula sentinel. Tracked in Add (*File).RecalcCell to persist a single formula cell's cached value #2307.
3. File.Recalc whole-workbook sweep built on RecalcCell, aggregating failures via errors.Join. Closes this issue.
4. RecalcOptions{Sheet, Ref} scoping for File.Recalc. Tracked in Add RecalcOptions{Sheet, Ref} to scope File.Recalc #2308.

Out of scope (tracked elsewhere)

• Quoted 3D references (SUM('A':'B'!ref)) — mis-tokenised by github.com/xuri/efp upstream of parseReference. Tracked in Support quoted 3D references of the form SUM('Sheet1':'SheetN'!A1) #2309.

Not needed (verified)

• _xlfn. prefix stripping already works (calc.go:210).
• ISOWEEKNUM and SHEET are registered and correct.

[gregkonush]

This proposal matches the boundary I would expect: keep the formula XML intact and only refresh the cached value/type.

A couple of edge cases that mattered when I implemented the same workflow in Bilig's XLSX path:

• Do not route the writeback through normal SetCell* APIs if they own formula removal. Cached-value writeback needs a smaller internal helper that updates <v> and the cell type while preserving <f>.
• Shared formulas need a regression fixture where both the master and dependent cells survive the cached-value refresh. It is easy to accidentally preserve the master formula but collapse the dependent cells.
• Store typed cache values, not only formatted display strings. Otherwise downstream readers can treat numbers as shared strings and aggregations change behavior.
• Treat unsupported formulas and external links as skipped/inconclusive rather than silently writing stale values. For a whole-workbook Recalc, a summary of evaluated/skipped/failed cells is often more useful than first-error abort.

For reference, the Node package I maintain for this same service workflow is @bilig/xlsx-formula-recalc: edit inputs, recalculate supported formulas, read proof cells, then export XLSX. Different language/runtime, but the useful invariant is the same here: formula text must remain reviewable while cached values become fresh.

Disclosure: I maintain Bilig, so treat this as implementation experience from a neighboring XLSX runtime, not a neutral recommendation.