docs(release): rewrite RELEASING and the release skill for the simpler flow

[datlechin]

Summary

• apple/RELEASING.md follows the flow from ci(release): check the tag's own CI runs instead of walking and polling #3584, ci: one CI workflow with a single required check #3675, ci: replace the changelog checker with towncrier and two small checks #3587 and ci(release-build): one MAC_RUNNER switch and a weekly hosted build #3589: 18,824 to 10,714 bytes.
• The train is four lines. Dropped the time window, holiday dates, ruleset pointers and the done v0.82 bridge paragraph.
• New Tagging section: version.py check runs on the local tag before git push, so a broken rule costs no number.
• Storage check runs before beta.1, before every stable, and on a pick that touches the storage paths. The critical question is a step at stable and patch.
• X.Y.0 step: the shipped detector must print false since the newest beta.
• Kept facts the gate and scripts still enforce: version bounds, the lead limit (6 plain lines, no # or - ), plugin retention, the Withdraw refusal, GA Xcode for app and plugins, publish-libs.sh --dry-run, libs-v1, the cert expiry warning, the release/0.82 gates.
• Fragments are curated in the release PR. The forward-port is an agent-run cherry-pick PR merged with --auto.
• retainFromPluginKit is checked only when TABLEPRO_PLUGINKIT_VERSION changes.
• New manual iOS section, tagged ios-vX.Y.Z, never v*.
• Reproducing a Release build moved to apple/CONTRIBUTING.md, with VERSION and BUILD exported (package.sh plugins needs BUILD). ci.md Runners has the MAC_RUNNER set and delete commands.
• Release skill: 7,028 to 4,162 bytes, six steps, agents run release commands when told. Model invocation stays on. It reads the shipped detector and checks the appcast and asset count after publish.
• Root RELEASING.md points iOS to apple/RELEASING.md. data-sync-storage.md points to the new Storage check section.
• .claude/rules/ci.md already sends iOS test flags to tests.md, Plans.

Test plan

• Ran every read-only command cited, against live state: tag list, ls-remote, regression issues, gh release view, gh run list (ci.yml with --status success and --commit, release-build.yml, release.yml), gh pr view --json mergeCommit, Dependabot alerts, gh pr list picks, the git grep heading check, version.py parse (including the beta and MINOR bounds).
• version.py check on local, unpushed tags in a scratch clone: a beta.1 reports only the missing CI push run; a second tag on the same commit reports one v* tag per commit.
• detect-changed-paths.sh --base v0.82.0 --head origin/main shipped prints true; the same head and base print false.
• Appcast grep on gh-pages prints 1 for 0.82.0; v0.82.0 has 30 assets, 4 plus 26 registry.py targets.
• Storage check over v0.82.0..origin/main lists the 4 expected commits.
• scripts/check-banned-words.sh --staged passes.