release: v2.2.0

[vyncint]

Milestone 2.2.0 (#50–#59), plus a deep documentation pass.

Deep test at this tree

check	result
just ci (fmt, clippy, test, docs, deny, schemas, pins, versions, skill)	green
just gate — real reconverge 0.6.0, cuda-oxide 26754ae5	3/3
MSRV 1.88 --locked	green
just termlens-cli (--ignored)	4/4
cargo-semver-checks --workspace vs 2.1.0, forced patch	8/8 crates clean
corpus prune --cc 8.6	93 clean, 8 refused, 0 tool errors
cargo oxide inspect × 6 kernels	all lower to PTX, one .visible .entry each
stress hunt, 100 iterations	12/12 — both OSes × 1/2/4/8/16 threads (run)

The 8 refusals are reduce-flip above one warp — the corpus's known flip, and the number that would move if the gate regressed.

The deep doc read found four false claims

Links and anchors are sound: a script walked all 22 markdown files and found 0 broken relative links and 0 broken anchors. The measured numbers hold — the five Spearman correlations match model-calibration.toml exactly, the documented cargo reconverge check --strict --message-format json --cc invocation matches runner.rs verbatim, and --cc 86 / sm_86 / 8.6 all resolve as claimed.

These did not hold:

1. --allow-unsafe was documented as if it were on tune. It is on stage; tune --allow-unsafe is an unexpected argument error. Verified, then corrected.
2. The README's CLI block omitted stage entirely — a real subcommand, listed in docs/ARCHITECTURE.md's crate table but nowhere a user would look. Added, with space --list and model --results.
3. just ci was described with a two-release-stale recipe list in three places — AGENTS.md, CONTRIBUTING.md, and ci.yml's own header. None mentioned docs, versions or skill.
4. docs/research-baseline.md said "the gate now pins 0.3.0" in the present tense. It pins 0.6.0. That row records the S0 measurement conditions and is deliberately frozen, so it now says which versions followed and points at the comparison sections.

Also: the mermaid diagram used \n for line breaks; <br/> is the form verified to render on GitHub, and \n in mermaid edge labels is unreliable.

Release mechanics

• workspace.package.version and the eight internal [workspace.dependencies] pins both at 2.2.0 — just versions enforces it, and stale pins would make cargo metadata refuse the next major.
• CHANGELOG [2.2.0] written from the ten issues, including the four that were wrong when filed — the miscounts are recorded because the shape of the miscount is the useful part.
• action/action.yml's version example bumped to 2.2.0. The floating @v2 tag stays; release.yml moves it.

After merge

Tag, publish, GitHub Release, install.yml, then baseline-version → 2.2.0 in a follow-up PR — per docs/RELEASING.md, which now spells out why the baseline moves after the publish and not before.

Signed-off-by: Vyncint Ng 115854244+vyncint@users.noreply.github.com