feat: NIP-23 long-form articles (kind:30023)

[tlongwell-block]

Summary

Accept kind:30023 (NIP-23 long-form content) in the relay's per-kind scope allowlist, advertise NIP-23 in NIP-11, and ensure global-only storage semantics.

Builds on PR 2 (NIP-33 parameterized replaceable events) — kind:30023 is in the 30000–39999 range, so replacement keyed by (pubkey, kind, d_tag) is handled automatically by the existing replace_parameterized_event() path.

Changes (4 files, ~430 insertions)

Production code (~15 LOC)

• kind.rs: KIND_LONG_FORM: u32 = 30023 constant + ALL_KINDS registration
• ingest.rs: KIND_LONG_FORM => Ok(Scope::MessagesWrite) in scope allowlist + extracted is_global_only_kind() helper to ensure channel_id = None for global-only kinds (strips stray h-tags at write time)
• nip11.rs: Added 23 to SUPPORTED_NIPS (extracted as module-level constant)

Tests (~400 LOC)

• ingest.rs: 7 unit tests — scope allowlist, channel scoping, global-only invariant, disjointness
• nip11.rs: 2 unit tests — NIP-23/33 advertised, sorted invariant
• e2e_long_form.rs: 5 E2E tests — acceptance, retrieval, stray h-tag global storage, NIP-33 replacement, stale-write protection

Design

• Scope: MessagesWrite — same as kind:1 (text notes). Articles are user-authored content.
• Storage: Global (channel_id = NULL). Author-owned, not channel-scoped.
• Replacement: NIP-33 via PR 2's replace_parameterized_event(). Keyed by (pubkey, kind:30023, d_tag).
• is_global_only_kind(): Extracted from inline matches!() in step 5b of the ingest pipeline. Makes the global-only invariant testable and documents the behavior.

Known Limitation

Stray h tags on global-only kinds remain on stored events (Nostr events are signed — tags can't be stripped without invalidating the signature). The read-path filter matching in filter.rs treats explicit h tags as authoritative, so a stray h tag can still match #h queries. This is pre-existing (kind:0, 1, 3 have the same behavior) and documented in code. Fix belongs in the filter layer as a follow-up.

Verification

• Unit tests: 156 relay + 42 core — all pass
• E2E tests: 5/5 against live relay
• Existing E2E suites: e2e_relay 26/27 (1 pre-existing nip05 failure), e2e_nostr_interop 15/15
• Live testing: kind:30023 publish, retrieve, NIP-33 replace, NIP-11 — all verified
• ACP agent harness: no regressions
• Pre-push hooks: fmt ✅ clippy ✅ tests ✅ desktop build ✅

Dependencies

• PR 2 (NIP-33 parameterized replaceable) must be merged first