[Rewrite P7] Cutover: release, migration run, parallel run, TS retirement #869

Parent: #860
Milestone: Phase 7: Cutover
Host: both · Expected cost: $600

Make nostream 4.0 the relay. The TS relay retires; the website stays.

• Re-run the Phase 0 audit against the NIP registry at cutover date — anything the protocol added since capture gets a disposition here (implement, defer with note, or skip) before the flag flips
• Release packaging: nostreamd container image (multi-stage, distroless, non-root), the release workflow ported from the TS one (GHCR publish on tag), changesets → release notes or the Rust-native equivalent (git-cliff) — decision recorded
• The compose/infra swap: nostream's deployment moves from the TS container + postgres to nostreamd + LMDB data dir; the pg-to-lmdb migration runs as the documented cutover step with the checksum gate from [Rewrite P3] Storage seam + LMDB backend (strfry-style) + PG→LMDB migration #864
• NIP-11 banner/version bump to 4.0.0; the supportedNips list updated to the Rust relay's audited reality
• Parallel-run window: both relays side by side on the production corpus snapshot — differential harness from [Rewrite P2] nostream-relay: wire handlers, strategies, rate limits #865 runs for 48h with zero divergences before DNS/traffic flips
• Rollback plan: keep the TS deployment + PG volume intact for 30 days post-cutover; rollback = re-point traffic, re-enable PG ingest (documented, tested once on staging)
• Post-cutover: retire the TS relay code to a legacy/ branch (kept for the rollback window), archive the postgres migrations, update the README/docs to the Rust architecture
• Actual-cost reconciliation: the factory manager fills every project item's Actual Cost USD from the factory spans; the total vs plan variance is posted to this issue

Acceptance: production serves from nostreamd+LMDB; differential harness green over the parallel window; the TS stack is retired to legacy/; costs reconciled on the board.