docs(planning): compact planning/, promote stranded truth to architecture/

[lesnik512]

Compacts planning/ and moves live facts into the truth home. Net -254 / +146 lines; the rationale the convention tells us to keep is preserved.

The finding that mattered

Two rules were live and load-bearing but written down nowhere:

• The profile README's DI table sorts alphabetically (modern-di first, that-depends last). I verified the table follows this today, but grep over architecture/ found no record of it. It was stranded inside a change file whose own frontmatter claimed the work hadn't happened yet.
• A repo may be listed before its package hits PyPI, at the cost of a temporarily 404ing Downloads badge.

Both now live in architecture/org-profile.md.

What changed

• architecture/community.md (new) — records what is actually true: GitHub Discussions on modern-di, that-depends, .github, and deliberately no chat rooms. I checked before writing: Telegram/Discord do not exist and are linked from nowhere, because the doctrine opens them "during launch week" and launch week never fired. Writing "the org runs Discord + Telegram" into the truth home would have made it lie.
• 2026-07-02.04-aiohttp-surfaces (108 → 26 lines) — said "Ready-to-fire", but every precondition is met (repo public, description/topics/homepage set, rows live on both surfaces). Retired the stale framing and deleted the embedded execution plan, which convention 2.0.0 bans from changes/. Kept as the why.
• deferred.md — the genuinely pending work now has revisit triggers: the coordinated launch, the launch-week chat rooms, the borrowed-audience program.
• Geometry trims — verbatim copies of shipped source (apparel_back, render_apparel, boosty_cover, _quantize_png, async_loop) and the org mark's coordinate/palette dumps replaced with pointers to the code that owns them. The palette table was reproduced exactly by tokens.py; the coordinates by geometry.py, pinned by test_geometry.py. Design rationale (the ~2px tail overlap, the textLength pinning, why FASTOCTREE) is kept.
• Ticked the verification boxes on the two merged changes (Metadata-consistency sweep: reconcile 12 profile blurbs #43, docs(profile): drop the type badge column from the templates table #44) — after re-verifying each claim, including a character-for-character comparison of all 19 profile blurbs against their GitHub descriptions.
• Fixed a broken changes/ link left behind by the 2.0 folder-to-file flatten.

What I deliberately did not touch

• docs-lockup-hero's CSS stays. It lives in other repos' docs/css/brand.css, so this change file is the only copy in this repo and the reference for rolling it out to new docs sites. Trimming it would have destroyed the sole record.
• The vendored convention (README.md, index.py, _templates/) — compacting it would diverge from lesnik512/planning-convention and break the APPLY.md update path.
• The 21 change files' rationale prose, and the launch playbook's copy.

Verification

• just check-planning → planning: OK
• just test → 149 passed
• Every pointer written names a symbol that exists (checked by grep against brand/build/); no broken relative links in planning/.