invites: NIP-86 membership claims share existing invitations

[FelineStateMachine]

Motivation

NIP-43 clients need standard management methods for caller-chosen membership invitations. bind.ws already has invites and a member tree, but exposes only its custom invite management API. This implements NIPs proposal #2408 at revision dcf5af03aacd5ca9c70c51448e32f60477f6ac34 (checked 2026-09-04).

Change

• listclaims([]) returns usable code strings; createclaim([code]) and deleteclaim([code]) return true on success.
• The existing invite table supplies expiry, uses, revocation and issuer attribution. All issuance paths share member quota/depth, permission checks and audit. Member listings filter by creator and liveness before the 200-code cap.
• Existing createinvite, listinvites, revokeinvite, NIP-43/NIP-29/HTTP joins and relay ownership claim retain their behavior. No database migration or broader permission redesign.
• Documentation pins the proposal and states the existing operational constraints: 4-64 ASCII code characters, three-day lifetime, unlimited uses until expiry/revocation, bounded listing, and member-scoped deletion. The proposal removes legacy kind-28935 invite requests; bind.ws never generated those replies.

This PR is independent of the parallel NIP-AD and NIP-9a changes, with no required merge order and no NIP-11 edits. README, HTTP docs and exposure tests may need ordinary conflict resolution when combined. The upstream proposal remains open; its unspecified expiry/code constraints follow existing bind.ws behavior.

Testing Completed

• npm run typecheck: passes TypeScript, generated-file, console, celld and template configuration checks.
• npm test: 44 files, 219 tests pass.
• Five new invite integration tests and a management exposure test cover authorization, chosen codes, audit, ownership separation, shared quotas, member isolation, filtered listing/cap, exact expiry boundary, exhaustion, revocation, bans and all join paths.
• Independent subagent production and final test review found no actionable defects.
• An initial full run hit an existing GRASP busy-response flake; its isolated 14-test rerun and the final full suite pass. New test timing assumptions were corrected before the final run.

Responsible AI disclosure

Implemented with Codex assistance, including subagent test work and independent review. Changes were checked locally with the project validation commands. No merge or deployment is included.