Migrate the remaining repos off the planning/ convention

Tracking issue for moving the rest of the org off the planning/ convention and onto
PR-body-as-spec. #50 did this repo; modern-di and faststream-outbox did themselves
earlier. This covers everyone else.

Where the org actually stands

Measured across all 28 non-archived repos, not estimated:

State	Count	Repos
Migrated — `CONTEXT.md` + `docs/adr/`, no `planning/`	3	`.github`, `modern-di`, `faststream-outbox`
Mid-migration — has both	1	`chat-app`
Not migrated — `planning/` only	21	the checklist below
On neither convention	3	`that-depends`, `fastapi-sqlalchemy-template`, `litestar-sqlalchemy-template`

Three of those are handled elsewhere and are not part of this issue's checklist:

chat-app — chat-app is half-migrated: the PR body is the spec, but planning/ is still a merge gate #70
that-depends — that-depends has no agent instructions file #69
fastapi-sqlalchemy-template and litestar-sqlalchemy-template are on neither
convention and have no planning/ tree, so there is nothing here to migrate. Whether
they should adopt one is undecided and untracked.

Why this is a programme, not a sweep

The .github migration (#50) was one PR but a session of judgment: ruling on 24 change
files, one decision record, a deferred list, and a 5-page architecture/ tree — each one
either derivable and dropped, enforceable and turned into a test, a rejected alternative
turned into an ADR, or real work turned into an issue. modern-di took five PRs
(modern-python/modern-di#424, #432, #433, #439, #449).

Per-repo issues should be spawned as work starts, not written upfront — 21 stale specs
would be worse than none.

The recipe, as actually performed twice

CONTEXT.md — the vocabulary. Seed from architecture/glossary.md where one exists;
otherwise author it, and audit every _Avoid_ entry against real usage before committing
(in .github that cut ten terms to seven — four rejected synonyms appeared nowhere in
the repo).
docs/adr/ — planning/decisions/* become NNNN-slug.md, each with a revisit
trigger. Also rescue rejected alternatives buried in change files and capability pages.
Deferred items become GitHub issues, self-contained, carrying their revisit triggers.
planning/changes/* are deleted. Git history is the record; route anything
load-bearing through the admission check first.
architecture/ is dropped. Enforceable claims become tests whose name is the claim,
with an INVARIANT: docstring naming what breaks it.
AGENTS.md gains the Workflow and Where-a-fact-goes sections and loses the lanes.
docs/agents/domain.md repoints to CONTEXT.md + docs/adr/.
justfile and CI drop check-planning / index. Add the offline link gate (Add the offline link gate to the 18 green repos, and unpublish the ADRs in the two that have them #66) in
the same PR — see below.
Do not touch .github/PULL_REQUEST_TEMPLATE.md. The org default is staying as the
generic form and is true under either convention, so a migrating repo needs no template
change. Template work is Delete the local PR templates; keep the generic org default #53, which runs independently of this issue.

The link gate rides along

#66 adopts a blocking offline link check in the 18 repos that are already green. Seven
repos are red, and all seven are red only inside planning/:

httpware 56, lite-bootstrap 8, compose2pod 4, semvertag 3, faststream-concurrent-aiokafka 3,
faststream-redis-timers 2, db-retry 1 — 77 broken links, 77 of 77 in planning/, zero in
any user-facing surface.

Deleting planning/ deletes all 77. So each migration PR should add the gate as its last
step and land green, rather than #66 excluding planning/ to work around history that is
about to be removed.

Checklist — the 21 not-migrated repos

chat-app (#70) and that-depends (#69) are tracked separately and are not on this list.

Out of scope

The org-default PR template and the local-copy removals (Delete the local PR templates; keep the generic org default #53).
Adopting the link gate in the 18 already-green repos (Add the offline link gate to the 18 green repos, and unpublish the ADRs in the two that have them #66).

The recipe is missing a step: `release.yml` depends on `planning/`

Found while migrating modern-di-pytest (the first off this checklist). Step 8 covers
justfile and CI's check-planning / index, but not the release workflow — and in the
_checks.yml-convention repos that workflow reads planning/ directly:

notes="planning/releases/${GITHUB_REF_NAME}.md"
if [ ! -f "$notes" ]; then
  echo "::error::Stable tag ${GITHUB_REF_NAME} has no curated release notes at ${notes}..."
  exit 1
fi

It does two things: hard-requires a curated notes file for every stable tag, and sources the
GitHub Release body from it. Delete planning/ while leaving that in place and the next
stable tag fails the release job — after just publish has already pushed to PyPI, since
the notes gate runs before publish but the body resolution runs after.

This is not one repo's quirk. Spot-checked locally:

lite-bootstrap — same three planning/releases/ references in release.yml
modern-di-faststream — same three
chat-app — different shape: main.yml invokes planning/index.py and planning/links.py
directly (relevant to chat-app is half-migrated: the PR body is the spec, but planning/ is still a merge gate #70)

There is already a precedent to copy. modern-di hit this during its own migration and
resolved it in 1ea74ee, "drop the planning/ directory and the curated-release-notes
convention" (modern-python/modern-di#449) — one of the five PRs this issue cites. Its
post-migration release.yml drops the notes gate and the body source, keeping only the
prerelease flag:

      - name: Publish GitHub Release
        uses: softprops/action-gh-release@v3
        with:
          generate_release_notes: true
          prerelease: ${{ steps.meta.outputs.prerelease }}
          draft: false

Suggested step 8b: release.yml drops the curated-notes gate and the
planning/releases/<tag>.md body source, matching modern-di post-1ea74ee. Worth calling
out explicitly that this retires a stated policy — curated notes were mandatory for stable
tags — so a migration PR should say so rather than let it ride in as a docs change. That also
means a migrating repo's PR is chore:, not docs:: it changes release behaviour.

One more note for whoever picks up the remaining repos: planning/releases/*.md are
byte-identical to the published GitHub Release bodies (checked on modern-di-pytest 3.0.1), so
deleting them loses nothing — the Releases are the record.