docs: close foundation roadmap for release

[rgehrsitz]

Completes the foundation release boundary after M8.10: M8 is marked complete and the independently gated dense commit allocation work moves to REX-M9. Adds reviewed v0.2.0-alpha operator notes, documents release-note preparation, and rejects malformed release tags before archive creation.

The supported release remains the serial durable daemon on a standalone Redis commit domain. Redis 7.0+ uses the default scripted transaction mode; Redis 6.2+ requires explicit WATCH mode. Shared-Redis concurrent workers and Redis Cluster remain outside the supported contract.

Validation

• GOCACHE=/tmp/rex-go-cache go test ./...
• GOCACHE=/tmp/rex-go-cache go test -race ./...
• GOCACHE=/tmp/rex-go-cache go vet ./...
• GOCACHE=/tmp/rex-go-cache go build ./...
• bash -n scripts/release/build-archives.sh
• invalid VERSION rejection returns exit 2
• VERSION=v0.2.0-alpha builds six release archives and all SHA-256 checks pass
• git diff --check

Release follow-up

After review and merge, create the annotated v0.2.0-alpha tag from the merged main commit. The tag-triggered workflow will publish the archives and generated GitHub release; verify its assets and apply the reviewed operator notes before declaring the release complete.

[rgehrsitz]

Review: docs — close foundation roadmap for release (M8 complete, M9, v0.2.0-alpha)

Verified on codex/rex-foundation-release (ba0f51e): go test ./... all packages pass, go vet/go build clean, git diff --check clean, bash -n clean, tree left clean (dist/ is gitignored). I also exercised the release mechanics end-to-end rather than trusting the script by reading. No blockers.

Release mechanics — verified by execution, not just inspection

• Tag → workflow → script chain is coherent for a prerelease. release.yml triggers on v* (matches v0.2.0-alpha), passes github.ref_name as VERSION, auto-adds --prerelease on -, and --verify-tag matches the annotated-tag procedure in RELEASING.md.
• Semver gate behaves correctly. I ran 11 VERSION cases: v0.2.0-alpha, v0.1.0, v1.10.3-rc.1, v2.0.0+build.5 accepted; bad, v1.2, v01.2.3 (leading zero), v1.2.3- (dangling hyphen), NOTAVERSION all rejected with exit 2, before any archive work. The regex is unquoted on the RHS (portable, safe under macOS bash 3.2), and exit 2 distinguishes bad input from build failure.
• Full archive build reproduces the PR's claim. A valid-VERSION run produced all six archives (4× .tar.gz, 2× .zip) and shasum -a 256 --check checksums.txt passes 6/6. Bonus observation: pushing a mistyped non-semver v* tag now yields a failed workflow run instead of a silently misnamed release — the intended hardening, just noting the failure mode is visible in Actions.

Docs consistency — checked against sources

• M8 Complete is justified. PRs #50 (0e24a3c) and #51 (c323ddb) are merged, so "M8.1–M8.10 are integrated" is factual; D15 now reads "Accepted for REX-M8.10 in PR #51", resolving the proposed-vs-integrated wording.
• M9 section faithfully transcribes D15. All eight bullets match the D15 gates I reviewed under PR #51: same 5,000-event fixtures, CV ≤ 20% + p95 max/min ≤ 1.20 control gates, 150KB/1,500-object targets, 1.10x / 0.90x dense gates, 0.98x/1.02x sparse guards, 5% Redis-CPU band, byte-identical formats, and revert-without-D13 falsification. No gate drift introduced in transcription.
• Release-notes contract claims verify. rexc v4–v7 + -legacy-v3 wirken as described per BYTECODE_COMPATIBILITY.md; Redis 7.0+/script vs 6.2+/explicit-WATCH matches the M8.9 operator docs; serial-daemon/standalone-domain scope matches D15; all four linked guides exist; v0.1.0-alpha tag exists as the stated baseline.

Nits (non-blocking)

1. Roadmap provenance convention drift. Earlier milestone rows carry merge commit hashes (e.g., M8.8 d01b5ad); the M8.9/M8.10 rows cite only PR numbers. Consider appending 0e24a3c and c323ddb for consistency — cheap, and this table is the project's audit trail.
2. Reviewed notes vs published notes can drift. The workflow publishes --generate-notes and never reads docs/RELEASE_NOTES_<tag>.md; the file is a manual review aid applied via step-5 edits. The procedure covers it, but the drift risk is structural — a future improvement could pass the reviewed file as the release body (or explicitly record why generated notes are preferred).
3. Follow-up sequencing is correct as written (tag from merged main, verify assets, apply notes before declaring complete) — just confirming the tag must be created after this PR merges, since the release boundary it declares doesn't exist on main yet.

Clean release PR: mechanics proven by execution, docs consistent with the evidence line from M8.8→M8.10, and M9 correctly staged as independently gated post-release work.

[rgehrsitz]

Addressed both review nits in 8702312:

• Added the M8.9 (0e24a3c) and M8.10 (c323ddb) merge hashes to the roadmap audit trail.
• Removed the reviewed-versus-published notes drift: the release workflow now requires docs/RELEASE_NOTES_<tag>.md and supplies it with --notes-file alongside --generate-notes, which prepends the reviewed operator guidance to GitHub's generated change notes. The release guide now documents and verifies that behavior.

Validation after the changes: workflow YAML parses, gh release create --help confirms supplied notes can precede generated notes, git diff --check passes, and GOCACHE=/tmp/rex-go-cache go test ./... passes.

[rgehrsitz]

README release audit completed in ebc19a7:

• documents all five packaged command-line programs, including rex_stressor;
• states the Redis 7.0+ scripted / Redis 6.2+ WATCH durable boundary, serial daemon contract, standalone commit domain, and unsupported Cluster/shared-Redis worker topologies;
• links reload and managed-partition ownership rollout guidance;
• aligns tag and reviewed/generated release-note behavior with the release workflow;
• replaces stale uppercase/group-shape prose with the actual recursive lowercase all/any grammar, v6 for, and v7 emit fields;
• removes the incorrect prohibition on colons in fact names and documents durable reserved names instead.

go test ./..., git diff --check, and a complete local README-link check pass.