docs(PROJECT_STATE): record the full drive-queue chain

[JDonaghy]

Why

The previous revision (PR #729, earlier today) named four per-backend pockets as "untracked residual — not yet issue-shaped." All four are now filed, plus a fifth it missed, and the whole platform-neutrality effort is on the drive queue. This replaces the residual section with the queue itself.

quadraui#666 ✅ ─┬─ vimcode#727 ▶ → #728 → #730 → #593 → #731 → #732 → #733
                 │                → #734 → #735 → #657 → #47
                 └─ quadraui#596 ▶ → #597 → vimcode#658

What changed

• The five pockets, with their issues — mouse routing Converge the two mouse routers — ~4,800 lines implementing the same precedence ladder twice #733, Msg bus Retire GTK's private enum Msg bus — 124 variants, 301 sites, a translation layer that only talks to itself #732, frame composition Converge frame composition — ~4,500 lines laying down the same surfaces in two hand-kept orders #735, orphan widget handles GTK: 22 Relm4-era widget handles are permanently None — the second orphan pocket #672 could not see (and why #723's fix cannot run) #731, and keyboard dispatch Converge keyboard dispatch — ~2,000 lines, and the TUI half's 19 'mirrors mod.rs:NNNN' pointers are all stale #734, the one the previous revision missed entirely. Its TUI half carries 19 mirrors mod.rs:NNNN comments and every one of those line references is now stale (mod.rs:1629 is a field initialiser today).
• Why the ">2 issues per repo" drive-queue caution doesn't apply here — the chains are strict, so at most one vimcode entry is ever tested-but-unmerged; there's no coord merge --revalidate drain to do. The two chains are in different repos and run in parallel.
• Honest sizing kept and sharpened — draining the chain cuts the backends ~38% to 14–15k production lines, which is not "thin event-to-engine wiring", and the remainder (rasteriser adapters, css.rs, window/CSD wiring, clipboard, font metrics) has not been enumerated. Says to re-run the sizing audit when Converge frame composition — ~4,500 lines laying down the same surfaces in two hand-kept orders #735 lands rather than assuming the chain finishes the job.
• Trust gate recorded as an accepted trade, not an open worry — Put vimcode on the oracle loop: promote gtk/render/tui_main into vimcode_core, then seal a tests/acceptance/ suite #657 is last because promoting gtk/render/tui_main first means rewriting every crate:: path across code GTK: 22 Relm4-era widget handles are permanently None — the second orphan pocket #672 could not see (and why #723's fix cannot run) #731–Converge frame composition — ~4,500 lines laying down the same surfaces in two hand-kept orders #735 then delete. Reversible with one re-chain. Also notes Put vimcode on the oracle loop: promote gtk/render/tui_main into vimcode_core, then seal a tests/acceptance/ suite #657's own 2026-08-10 bug-fix freeze has not been honoured for three weeks.
• Milestone hygiene — Extend Lua plugin API to expose quadraui primitives #146 moved out of Release v0.3.4 — bug fixes #7 (an addition among deletions); Native macOS GUI — thin wrapper over quadraui::macos::run_with_shell (re-scoped: no Core Graphics work) #47 into fix: regenerate Flatpak vendor sources for v0.3.2 #5 per GOALS.md's own definition of that milestone. Release v0.3.4 — bug fixes #7 is now 10 open / 29 closed, and all ten are queued.
• quadraui#465 closed 2026-08-31 and is present at our pinned rev 69fd9cdd (macos/shell_runner.rs:24), so the macOS gate is cleared and Native macOS GUI — thin wrapper over quadraui::macos::run_with_shell (re-scoped: no Core Graphics work) #47 needs no pin bump.
• Corrects #7 stands at 29 closed / 4 open, left stale by this morning's PR.

Testing

Docs-only (pure .md), exempt from the black-box driver bar per CLAUDE.md.

🤖 Generated with Claude Code

https://claude.ai/code/session_016N1UjBkNdxrm4YEd6oYJAz