Code-split the spreadsheet engine out of the document WASM bundle #30

Parent track: #21 (bundle-size ratchet-down). Split out per the 2026-07-10 investigation there — this is the one real bundle-reduction lever found, but it's a multi-day architectural project, not a gate chore, so it lives on its own.

Summary

The spreadsheet engine + its UI compile into the same single *_bg.wasm as the document editor and load on every page, even pure-document sessions that never open a sheet. Deferring it behind a dynamic load is the largest available WASM reduction.

Size (rough, from the #21 twiggy investigation)

• Engine (src/spreadsheet/{eval,functions,parser,pivot,stats,translate}.rs) + UI (src/components/spreadsheet_view.rs) = 294,746 B raw pre-opt = 5.3% of real code / ~25% of our own app code.
• Proportional estimate on the shipped baseline (~1,712,714 B gzip): ~85–90 KB gzipped saved on non-spreadsheet page loads.
• Heavily caveated: linear pre-opt→post-opt-gzip scaling is not exact; treat as order-of-magnitude, not a promise. A precise number needs wasm-opt with names preserved through the pipeline (not currently installed).
• Single biggest function in the engine: spreadsheet::functions::call_function = 124,301 B (a large match over every sheet function) — a secondary, self-contained target if the split proves too heavy.

Why it's not a quick change (separability blocker)

• src/pages/document.rs imports SpreadsheetView unconditionally (use crate::components::spreadsheet_view::SpreadsheetView;) and branches on doc_type at runtime inside one page component — there is no module boundary isolating spreadsheet code from the document page today.
• Genuine lazy-loading requires either:
  1. A second wasm-bindgen target compiled to its own .wasm + JS-side import() glue, plus a way to bridge Leptos reactive state across two wasm instances; or
  2. Restructuring pages::document.rs to defer the SpreadsheetView import behind a dynamic-loading shim.
• Neither exists in the current Trunk / Leptos-CSR setup. This is the bulk of the work.

Don't forget the CSS

spreadsheet.css (~37 KB, frontend/dist/spreadsheet-*.css) is loaded unconditionally via a static <link> in index.html — outside the WASM gate. Splitting the WASM without also deferring this CSS only captures part of the win.

Severity

S4 — no user-facing issue; a bundle-weight optimization. Deferred: high effort, medium-uncertain payoff, and even a perfect split leaves the bundle at ~1.61 MB (Leptos framework internals are ~34.5% of the bundle and structurally dominate — see #21). Pick up when someone has multi-day appetite for the CSR code-splitting infra, not under gate pressure.

Suspected area

• frontend/src/pages/document.rs — the unconditional import + runtime doc_type branch
• frontend/src/components/spreadsheet_view.rs, frontend/src/spreadsheet/ — the code to split
• frontend/index.html — static spreadsheet CSS <link>
• frontend/Trunk.toml / build config — where a second wasm target / dynamic import would be wired