ci(release): add MCP release candidate dry-run workflow

[JSONbored]

Parent roadmap: #82

Background

MCP releases need repeatable preflight checks before a tag is pushed. The package is public, and release mistakes are harder to undo than normal PR mistakes.

Goal

Add a release-candidate workflow that validates the package, changelog, tarball, CLI smoke, and provenance readiness without publishing.

Current Behavior

Release checks exist across scripts/workflows, but there is no single dry-run path that maintainers can run before creating a release tag.

Desired Behavior

A maintainer can run one manual workflow or local command set and know whether an MCP release candidate is safe to tag.

Implementation Requirements

• Add or wire a release-candidate check that does not publish.
• Verify package version and intended tag format.
• Verify MCP changelog has a real target version section.
• Run package tarball allowlist and secret scan.
• Run packed-package CLI smoke.
• Verify trusted publishing workflow configuration stays tokenless.

Public/Private Output Boundaries

Dry-run logs must not print tokens, npm credentials, GitHub auth details, local paths, or environment dumps.

Acceptance Criteria

• Maintainers can run the dry-run before tagging.
• The workflow fails on malformed tag/version assumptions.
• The workflow fails if tarball contents include secrets or unintended files.
• The workflow emits clear next steps without requiring a publish attempt.

Testing Requirements

• npm run test:ci must pass.
• Global coverage must remain at or above 97% for lines, statements, functions, and branches.
• Aim for 98%+ branch coverage locally to avoid CI variance.
• Add tests for every new branch, fallback path, sanitizer rule, and regression.
• Add invariant/property-style tests when behavior depends on sorting, gating, scoring, queue pressure, source-upload safety, public/private boundaries, or upstream drift.
• Public GitHub output must be tested against forbidden language: wallet, hotkey, raw trust score, payout, reward estimate, farming, private reviewability, and public score estimate.
• MCP/local tooling must prove source contents are not uploaded.

Additional Test Scenarios

• Success fixture with matching package/version/changelog.
• Failure fixture for missing changelog section.
• Failure fixture for unexpected tarball file.
• Failure fixture for tag/package mismatch.