Canonicalize table addressing across APIs (#450)

[JSv4]

Closes #450

Stacked on #474 because tracked block-SDT deletion (#473) is the structural foundation for the remaining epic work.

What changed

• makes canonical tc anchors the single cell-oriented mutation contract, with a documented legacy paragraph-anchor migration shim
• exposes explicit table/row/column/cell metadata and bidirectional cell-anchor/coordinate resolution across core, WASM/npm, Python/stdio, and MCP
• centralizes gridBefore/gridAfter, gridSpan, vertical-merge, and nested-table geometry in one shared model
• returns deterministic retained/added/invalidated structural anchor mappings for row/column edits
• keeps incomplete tblGrid inspection read-only via deterministic virtual column identities, materializing real columns only inside mutations
• fixes the related DocumentStructure coordinate/anchor defects from DocumentStructure table coordinates are untested and computed incorrectly #471 and updates editor/tool documentation
• closes the IR projection vocabulary over canonical col anchors so table reads preserve oracle AnchorIndex parity instead of throwing

Verification

• focused table/addressing/MCP .NET aggregate: 108/108
• independent canonical-addressing rerun: 8/8
• IR column-anchor regression and token roundtrip: 2/2
• nearby IR reader/value/table-addressing suite: 38/38
• corrected [Agent] Normalize table, row, column, and cell addressing #450-through-[Agent] Extend native revision handling to structural edit families #455 integration matrix: 197/197
• Python table E2E: 1/1
• npm production and test typechecks
• core, MCP, Python host, and WASM builds (0 errors)
• OpenXmlValidator coverage, save/reopen anchor stability, nested-table isolation, ragged grids, spans, vertical merges, headers, revision-bearing tables, and IR/Markdown projection-anchor parity