release: 4.0.0 as the base line + docs/vocabulary migration

[xywang68]

Cuts 4.0.0 as the base line and finishes the vocabulary migration that was stranded when #170 was squash-merged.

The two lines

Line	Image	Status
3.0.0 — framework	xyteam/autobdd-framework:3.0.0 (+ the xyteam/autobdd alias)	untouched. Tag, release and image stay exactly as shipped — base + Chrome + WebdriverIO + Cucumber.
4.0.0 — base	xyteam/autobdd-base:4.0.0	this release. A docker-runnable GUI (Xvfb + openbox + x11vnc) with the Oculix screen engine behind one CLI, so higher-level tools — a framework, a script, an agent — drive it. No browser, no runner, no BDD layer.

v3.0.0 is a real tag in this repo and nothing here rewrites it.

The version is now real, not implied

• .docker/autobdd-base.dockerfile: ARG AUTOBDD_VERSION=4.0.0, recorded as version= in /etc/autobdd-versions. The framework image already selects its base with the same arg (FROM xyteam/autobdd-base:${AUTOBDD_VERSION}), so the two lines stay coherent and re-baselining the framework onto 4.0.0 is a later, separate change.
• autobdd --version reports it:

autobdd — AutoBDD base image
version: 4.0.0
image  : built 2026-09-22T21:01:45Z
oculix : oculixapi-4.0.0-linux.jar
node   : v24.21.0

• package.json / package-lock.json: 3.0.0 → 4.0.0.
• The suite gates the release: provenance asserts version=4.0.0 in /etc/autobdd-versions, and version asserts --version reports it. A wrong or missing stamp fails CI.

Bug found while wiring it

The provenance stamp's oculix= field pointed at /opt/autobdd/third_party/xysikulixapi/lib — a path that stopped existing when the seam became in-tree earlier in this PR series, so the field was silently empty. Now reads /opt/autobdd/seam/lib and is populated.

Also in this PR: the stranded documentation audit

Squash merges break ancestry, which is why this is a second PR: git merge-base --is-ancestor reports every original commit as missing after a squash, so the tree diff is the proof — master differed from the branch in exactly these files, and git diff bd12649 HEAD here is empty (byte-identical to the tree verified at 133 passed / 0 failed).

• README.md — the seam section documented the v1 arguments as current; now the canonical table and examples, with docs/CONTRACT.md as the source of truth so they cannot drift again. Real run lines in the sample output, the duplicated Basic usage block removed, and the opening rewritten for the 4.0.0 re-scope.
• docs/CONTRACT.md — the v1 table was dangling under a "Raw arguments" heading as if current; retitled Raw argument reference (v1 — deprecated) pointing at the §2b mapping.
• framework/libs/screen_session.js — the last v1 emitter. It built findTargetImage --imagePath=… --imageAction=…, so every framework call would warn on stderr seven times. Now emits the front door with canonical flags. Reviewer-relevant subtlety: textHint was a regex and --match-text is literal by default, so --match-regex is passed to preserve existing semantics exactly.

autobdd find-target --match-image=/tmp/…/hello.png --min-score=0.8 --max-score=1 \
  --match-text="HELLO" --match-regex --wait=5s --hover --click --limit=1
→ {"name":"hello.png","score":1,"clicked":"object"}      stderr: (empty)

• seam/ + suite: known-gap wording now says --box (what a reader types); the help feature asserts the canonical flag and the deprecated v1 list, so §2b cannot silently vanish; group I's discovery features invoke the surface under test rather than a hardcoded binary, so the log matches the banner.

Verified

Check	Result
autobdd --version	reports version: 4.0.0
Base suite	133 passed, 0 failed
one.sh I (discovery group)	18 passed
Framework e2e suite	not run — needs the L2 image with Chrome. That consumer is verified at the command-string level plus the engine's v2/legacy coverage, and re-baselining the framework onto 4.0.0 should not happen until it passes.

No v1 flag name remains in any documentation except the §2b deprecation table and CHANGELOG.md's historical notes.