docs: align Core 3.0 beta migration and consumer witnesses

[morozsm]

7 code/docs + 0 regenerated baselines = 7 files

Migrate 2.11.1 consumers to the Core 3.0 beta contracts and retain executable HTTP/state examples alongside the guide. The guide covers all 21 frozen compatibility rows, receiver observation metadata, intentional legacy HTTP PTT rejection, canonical TX ownership/admission/completion/OFF, integer-centiHz setters, extension host 2, custom tone tables, timed CLI PTT and raw rigctld errors. Package and editable lock metadata become 3.0.0b1, an unreleased candidate.

MOR-2338 and MOR-2340, with documentation ownership under MOR-1101 and frozen decisions in MOR-2336.

Merge dependency: extension-host PR #3185 MUST land before this documentation merges. This PR targets main and does not copy extension-host commits. Python compatibility already landed in #3183. Astra is the merge operator; independent review must assess this final head before the guarded merge.

The seven-file/670-changed-line soft threshold is justified by one compatibility acceptance unit: the guide and public contract, narrowly aligned host-version policy, reconciled changelog, two matching package-version files, and the frozen HTTP/state consumer witness. The hard 1000-line limit is not exceeded. Historical published changelog entries and v1-to-v2 migration guidance remain; superseded unreleased TX lease/watchdog claims are replaced by the canonical migration contract.

Validation:

• Reused MOR-2338 Mini evidence: 11 passed, fake radio/authority, real ephemeral TCP HTTP parsing. WS evidence covers production dispatch and JSON serialization through fake send_text, not network upgrade/frame transport or RF. The witness was cherry-picked unchanged from 1e9d1bff6bcb60c2c010f2ee66532b514086b51b.
• Doc citation and relative-link gates PASS; both baseline-growth checks PASS against origin/main; git diff --check PASS. Initial link failure exposed the root changelog symlink; the four new migration links now use the canonical documentation URL.
• Static TOML validation confirmed matching 3.0.0b1 and no other semantic metadata changes. All 21 matrix IDs and the exact file lease were checked. No local product suite rerun; this Ready PR receives its natural applicable CI.

Installed-wheel smoke (MOR-2341), hardware, browser/operator and final owner acceptance remain pending release gates. HTTP 202 is admission without settlement; unobserved SUB notch can serialize zero with missing/unknown metadata; an extension transport false can leave an offline command queued. No old TX lifecycle emulation, full compatibility, successful publication or Pro release is claimed. No tag or release is created.

Review corrections: removed the duplicate historical host-global spelling from the public API surface and linked the migration guide; both unreleased compare references use the existing v2.11.1...HEAD range (GitHub compare API resolves successfully); policy section 6 now states host API 2 / interface V2 and explicit manifest host_api 2.0 while preserving schema version 1. Citation, link, both baseline-growth, rebrand and diff checks PASS on the correction. No product suite was repeated.

[morozsm]

Agent Review: BLOCKED c7b5e36

Independent review of all six changed files (591 additions / 71 deletions), including all new prose and the 11 parameterized consumer cases. Two bounded documentation corrections are required:

1. [P2] Keep the new API-surface text compatible with the rebrand gate. docs/api/public-api-surface.md:228-229 adds window.icomLanExtensionHost in a file outside the intentional historical/migration allowlist. The exact-head grep-gate failed on this path (run); this is a concrete new-content failure, not pending CI. Prefer describing the deprecated naming alias here and linking to the already-allowlisted migration guide, which retains the exact spelling. Alternatively an intentional allowlist entry needs its own justified scope. Verify the resulting changed prose with .github/scripts/check-rebrand-allowlist.sh and the new head's natural gate.
2. [P2] Do not make unreleased changelog navigation depend on a nonexistent tag. docs/CHANGELOG.md:2565-2566 now points both comparison references through v3.0.0b1, although this PR explicitly prepares metadata without creating that tag. git ls-remote --tags origin refs/tags/v3.0.0b1 returns no tag and gh api repos/rigplane/rigplane-core/compare/v2.11.1...v3.0.0b1 returns HTTP 404. Keep these unreleased comparisons on the existing v2.11.1...HEAD range (or another existing, truthful ref), and change them to the beta tag only when publication actually creates it. Verify both changed comparison targets resolve; do not create a tag as a workaround.

The reviewed migration semantics otherwise agree with the inspected production contracts: centiHz and duplicate-keyword rejection plus released profile fallback expressions from merged #3183; receiver-scoped values and unknown metadata; four rejected legacy HTTP PTT forms; stable-owner WS admission and owner disconnect; distinct intentional latched TRANSMIT / ForceOff admission; strict extension-host 2 contract and honest transport boolean from dependency #3185 at ae2d187. This is source-contract verification for the documentation, not a second code review of #3185. That PR MUST merge before this documentation PR.

The new tests use production ephemeral-loopback HTTP parsing and dispatch/JSON serialization with existing fake authority/radio helpers. Their assertions cover the advertised receiver, rejection, ownership and admission boundaries. Their WS scope does not cover network upgrade/frame transport. Reused builder evidence reports 11 Mini PASS; no suite, build, installed candidate, live server or hardware execution was repeated or claimed by this review. Version edits are limited to matching 3.0.0b1 metadata. Published historical changelog sections remain intact; the removed development claims were in unreleased material.

CI observed during review, independently of this code/prose verdict: quick SUCCESS; consumer-contracts and state-types SUCCESS; grep-gate FAILURE as above; Agent Review Gate and v2 observation had no PASS yet; quick-v2 observation PENDING. The coordinator must recheck all required contexts at the corrected exact head before merge.

[morozsm]

Agent Review: PASS 958a9f1

Independent corrected-head review complete. Reviewed the entire delta from c7b5e36 to this head and retained the prior review evidence for unchanged content. Tree: 7361d05dd890d74ccfc1b7483fec45ed9e11cd68.

Both prior BLOCKED findings are resolved: the public API page describes the deprecated naming alias and links to the migration guide instead of introducing a rebrand leak; both unreleased changelog comparisons now use the existing v2.11.1...HEAD range. I independently queried that compare API successfully (status ahead). The additional leased change in docs/architecture/open-core-policy.md section 6 accurately aligns numeric host 2, LocalExtensionHostApiV2, explicit host_api "2.0", rejection of omitted/old/unsupported declarations, and unchanged manifest schema version 1 with the source contract already inspected. No source, test, or package metadata changes occurred in this correction.

No remaining code/prose findings or mandatory body corrections. The seven-file/670-line scope rationale is recorded in the PR body and stays below the hard ceiling. Existing 11-test Mini witness evidence is reused; no suite, installed candidate, live server or hardware execution was repeated. Installed/browser/hardware/owner acceptance and publication remain separate pending gates.

CI observed without waiting: rebrand grep-gate SUCCESS; quick QUEUED; consumer-contracts and state-types IN_PROGRESS; Agent Review Gate was awaiting this exact-head verdict. The coordinator must check required contexts at this head before merging and must land extension-host dependency #3185 first. This PASS is the independent code/prose verdict, not permission to bypass either gate.

[morozsm]

The failed quick run 33932630652 stopped in the unchanged test_bcd_encoding_throughput wall-clock benchmark: 22,620 operations/s against a 50,000 floor over a 100 ms sample. It reported 15,547 passed and one failed test.

Comparison of c7b5e36 and 958a9f1 found identical blobs for the profiling test, bcd_encode, and its public import; the intervening commit changes three documentation files. An isolated Mini run of that single test at 958a9f1, using Python 3.11 and verified checkout import provenance, passed at 965,142 operations/s. The preceding PR head's quick run also passed; its exact throughput was not retained. Historical runner load cannot be reconstructed from this evidence.

One rerun of the failed quick job is requested on the unchanged reviewed head. The performance threshold remains unchanged. If the same failure repeats, retries stop and the existing parked benchmark proposal #2899 will be reassessed against the observed failure. No source change or new review cycle is required for this rerun.