ci(release): align beta branch config with main's canonical form

[DrumRobot]

Summary

beta is structurally unable to publish. semantic-release reads the triggered branch's own .releaserc.json/.releaserc-npm.json, and both production and beta still list {name: "main", channel: "next", prerelease: "next"} as the prerelease entry — while main's own config already reads {name: "beta", channel: "beta", prerelease: "beta"}. Evidence: release job run 29759934202 logged configured to only publish from production, main.

This is Phase 1 of plan-npm-beta-config-restructure.md (Option A, user-approved). Aligns production/beta's config with main's already-correct form. No workflow trigger changes needed (push: branches: [beta, production] stays as-is).

This PR is opened as a draft against beta first (not production) per the "beta before production, always" rule — using this PR's CI preview job as the Phase 0 dry-run verification, since local npx semantic-release --dry-run repeatedly failed on an unresolved npm/pnpm environment issue (see plan Progress Checklist for the failed local repro attempts).

What to check in the preview job

• npm-semantic-release.yml preview job output no longer says "configured to only publish from production, main"
• Preview reports a computed next version with a -beta.N suffix
• vsx-semantic-release.yml preview shows the equivalent for the vscode line

Do not merge without explicit approval — merging triggers the beta-vscode/beta-npm environment gate, which is a real publish path.

Test plan

• Verified via git show origin/{main,production,beta}:.releaserc*.json that the asymmetry described above is real (not stale-doc confusion — cross-checked against .claude/rules/branch-source-matrix.md, which is a known-stale doc already tracked in Issue .claude/rules/branch-source-matrix.md describes a superseded release-branch model #208)
• Preview job Job Summary confirms a computed -beta.N version (post-push, pending CI)

[DrumRobot]

AI Review Summary

Reviewer matrix: CodeRabbit (review disabled for this base branch — repo policy) + Copilot (no active subscription on the acting account, no existing review) → Internal Code Review (auto-fallback, code-reviewer subagent).

Internal Code Review findings

#	Severity	Finding	Verdict
1	—	JSON syntax/structure preserved in both files, diff confined to the single branches line	✅ Verified — no issue
2	—	No side effect on the production entry in either config (byte-identical before/after)	✅ Verified — no issue
3	—	tagFormat / plugin config / asset lists untouched in both files	✅ Verified — no issue
4	—	No risk to the production stable-release path; workflow trigger unchanged	✅ Verified — no issue
5	Important (context, not a diff bug)	.claude/rules/branch-source-matrix.md describes beta as retired, contradicting this PR's premise	Acknowledged — see below

On finding #5: this is .claude/rules/branch-source-matrix.md, already confirmed stale in a prior investigation — it was written 2026-07-19 and superseded by PR #205 (merged 2026-07-20, "restructure release branch strategy to separate release commits"). Primary-source verification (git show origin/{main,production,beta}:.releaserc*.json) confirms the asymmetry this PR fixes is real: main's config already reads {name:"beta",channel:"beta",prerelease:"beta"}, while production/beta still read the stale {name:"main",channel:"next",prerelease:"next"}. The doc itself is tracked separately for correction: #208.

Verdict

0 Critical / 0 Important (diff-correctness) / 0 Minor. The 2-line config change is correct, minimal, and isolated. Real CI signal (a semantic-release dry-run against this branch's actual config) requires an actual push to beta — the workflow triggers only on push, not pull_request. Recommend merging to get that signal, with the understanding that the preview job runs unconstrained (no environment gate) and only the separate beta-npm/beta-vscode environment approval can trigger an actual publish.

Formal Review: not posted. Self-authored PR (author = acting account DrumRobot); the es6kr org's review-account counterpart (daegunjhy) shares the same commit-author email (drumrobot43@gmail.com), which GitHub treats as a self-approval attempt regardless of which account posts it — per self-review-account-switch.md condition 4, this falls back to issue-comment Summary only.