[Repo Assist] Add Excel docs page; optimize Frame.stack and Frame.unstack

[github-actions]

🤖 This is an automated pull request from Repo Assist, an AI assistant for this repository.

---

Summary

Two improvements in one PR (Tasks 10 and 8 from this run):

---

Task 10 — docs/excel.fsx: Excel integration documentation page

Adds a comprehensive documentation page covering both Excel packages:

Package	Coverage
Deedle.Excel.Reader	Cross-platform reading of .xlsx/.xls files via ExcelDataReader
Deedle.Excel	Live read/write against a running Excel instance (Windows only)

Sections

• Package setup — NuGet install instructions and open statements
• Reading the first worksheet — readExcel
• Reading by sheet name — readExcelSheet
• Reading by sheet index — readExcelSheetByIndex
• Listing worksheets — sheetNames
• Working with the resulting frame — typed column access, indexing, merging sheets
• Type coercion and missing values — how obj columns are coerced; empty cells as missing
• Error handling — exceptions thrown for unknown sheets; try/with pattern
• Writing to Excel (Windows-only) — brief overview of Deedle.Excel live-write API

Two small sample data files are added to docs/data/ (sales.xlsx and quarterly.xlsx) to support live evaluation in fsdocs. A link to excel.html is added to docs/index.fsx.

---

Task 8 — Performance: Frame.stack and Frame.unstack

Frame.stack

Before: the nested for rowKey in rows do for colKey in cols do loop called frame.ColumnIndex.Locate(colKey) and frame.RowIndex.Locate(rowKey) on every cell — O(rows × cols) total dictionary lookups for each axis.

After: precompute per-column data vectors (once per column) and per-row addresses (once per row) before entering the nested loop. The inner loop now only calls vec.GetObject(rowAddr) — no index lookups.

Operation	Before	After
Column address lookups	rows × cols	cols
Row address lookups	rows × cols	rows

Frame.unstack

Before: iterated frame.RowIndex.Keys twice — once to collect unique fst keys and once for snd keys (Seq.map fst |> Seq.distinct |> Array.ofSeq).

After: single pass using HashSet<'R1> and HashSet<'R2> with corresponding ResizeArrays, maintaining first-occurrence order. Keys are iterated once.

---

Test Status

Passed!  - Failed: 0, Passed: 703, Skipped: 0, Total: 703

All 703 tests pass. Semantics are unchanged — same outputs, fewer allocations and dictionary lookups.

Generated by 🌈 Repo Assist, see workflow run. Learn more.

To install this agentic workflow, run

gh aw add githubnext/agentics/workflows/repo-assist.md@7ee2b60744abf71b985bead4599640f165edcd93