Declaring a table in the streaming workbook writer

Cluster: streaming

Scenario

A user writes a large spreadsheet through the streaming workbook writer to keep memory bounded, and wants to define an Excel table over the rows they streamed out, for filtering, sorting, banded styling and structured references. In the buffered path a worksheet exposes an add-table capability, but the streamed worksheet either lacks the method or throws at runtime when it is called. Worse, the worksheet type still advertises the add-table method, so callers discover the gap only at runtime with a "not a function" error rather than at compile time. Because a table's XML footprint, its column and header definition and cell range, is known once the row range is committed, declaring a table in the streaming path without buffering the whole sheet should be possible.

Spec note, not a corpus case: the feature does not exist yet, so there is no behavior to assert. The durable value is the streaming table declaration model and the honest-type requirement.

Desired behavior

• The streaming writer supports declaring a table over a range of streamed rows, producing the same table part XML, header row, autoFilter, table style, and worksheet-to-table relationship the buffered writer produces for an equivalent table.
• The public type surface is honest: a streaming worksheet must NOT advertise a table-adding method it cannot fulfill. Either the method exists and works in streaming mode, or the streaming worksheet type omits it so misuse is a compile-time error, never a silent runtime throw against a signature that claims support.
• The table's declared range stays consistent with the actual written extent, the header labels match the header cells, and the table never silently overwrites or reflows already-flushed cells, since the header row and any totals row are part of the streamed content.

Prior art / constraints

Streamed rows are flushed incrementally and cannot be re-read, so the declaration model differs from the buffered one. Two workable shapes: declare the table up front (name, header columns, style, totals intent) then stream the data rows into it, finalizing the cell range on commit from the row count; or stream rows first, then attach a table over an explicit committed range before finalizing the sheet.

Open questions

• Require the header row in the table declaration, as authoritative column names, or infer it from the first streamed row?
• How are totals rows handled when the row count is known only at commit time?
• When the declared table range and the streamed row range disagree: hard error, or clamp?
• Should the same declarative table spec the buffered round-trip uses also drive the streaming writer so both paths share one description?

Related: streaming-write-add-image, streaming-writer-worksheet-splice-rows-columns, table-headerless-omits-autofilter, existing-table-roundtrip-fidelity.