fix(changeset): reject type-none release-plan entries

[PaulNewling]

Problem

The block-release gate passes when it should fail, so a PR can merge green and ship nothing.

changeset status puts every dependent of a bumped package into .releases, at "type": "none" when no version change is due. Both changeset actions read that array by name only, so a none entry counted as a bump.

In a block repo the private model/ui/workflow siblings are versioned — no block repo sets changesets' privatePackages, and the default versions private packages. A sibling-only changeset therefore releases the siblings and lists the block as their dependent at none:

@platforma-open/milaboratories.top-antibodies.ui      patch
@platforma-open/milaboratories.top-antibodies.model   patch
@platforma-open/milaboratories.top-antibodies         none   ← read as a bump

require-package-bump accepted that. platforma-open/antibody-tcr-lead-selection#181 merged with a green check for changesets and released nothing; #183 then had to force the release with an empty changeset. 18 block repos opt into this gate today, so the same hole is live in all of them.

check-coverage carried the identical defect: a none entry satisfied the coverage requirement for a package whose own files the PR edited.

Fix

Both scripts now filter .type != "none".

Why the existing test didn't catch it

require-package-bump's suite already had "fails when only a sibling changeset is present", and it was green. The fixture pinned "privatePackages": false, which drops private packages from the release plan and suppresses the cascade entirely — so the test passed without ever producing the entry the gate has to reject.

Removing that key makes the fixture match a real block repo, and turns the test red without the script fix. check-coverage's fixture gains a publishable pkg-lib/pkg-app pair on a workspace:* devDependency edge for the same reason.

Each suite also asserts the none entry directly, so a future fixture edit that re-suppresses it fails loudly instead of silently disarming the gate.

CI gap

Nothing ran the require-package-bump suite. 0-test-changeset-coverage.yaml was scoped to actions/changeset/check-coverage/**. It becomes 0-test-changeset-actions.yaml: a two-leg matrix over actions/changeset/**, covering both actions. Nothing referenced the old filename.

Verification

Both suites run locally, each proven red before the fix and green after:

Suite	Result
require-package-bump	11/11
check-coverage	23/23

The none cascade was first reproduced outside the suites, against antibody-tcr-lead-selection#181's exact tree, which is where the output above comes from.

Notes for the reviewer

• No file in this diff contains a milaboratory/github-ci self-ref, so the @v4 / @v4-beta flip does not apply to any of it.
• v4-beta was 15 commits behind v4 and did not contain actions/changeset/require-package-bump/ at all — it reached v4 directly via MILAB-6648: require a changeset for every edited package, in one gate #198. fix-beta.sh was run to re-sync before this branch was cut.