spec: add the core/room domain, unifying rooms and DMs

[Mearman]

Closes #31.

Adds core/room: one domain covering both named, persistent rooms and two-party DMs, gated by a single room:member capability for the four ordinary participation verbs (room.send/room.read/room.leave/room.members), mirroring exec:pty's own one-resource-many-verbs shape. room.join/room.invite are deliberately ungated -- no token field at all -- since a room with no prior relationship has nothing to gate first contact on; approval mints a token on the spot and returns it as an extension field on the existing manage-ok response.

Membership proof is rooted in the room-path's own shape: an owner-named room (<owner-hex>/<local-name>) roots trust at the owner named in the path, but a DM path (the sorted pair of the two participants' hex device-ids) roots trust at the verifier itself -- if either named party were an acceptable root, a sender could self-issue a token and message a stranger unsolicited.

Two things found and fixed while building this, not left as known gaps:

• Unbounded delegation. Any bearer of room:member could mint a narrower child carrying the identical verb (narrowing preserves the capability), so any member could transitively admit anyone to any depth, and kicking one member's grant would take down their entire invited subtree since only a token's own issuer may revoke it. Fixed with a new ? delegations-remaining: uint claim on token-claims -- absent means unbounded (every existing token keeps its current meaning), present means a child's value must be strictly less than its parent's. General, not room-specific: exec:pty has the identical latent bug and gets the fix for free.
• A pre-existing cddl.js bug, unrelated to this domain's own design. The vendored CDDL lexer never processed backslash escapes inside string literals at all, so a .regexp pattern needing an escaped backslash (dm-room-path's \\+, and the already-shipped namespaced-domain-id's \\.) generated a JS regex with an extra literal backslash, silently matching the wrong thing. Fixed at the source: webdriverio/cddl#91/#92 upstream, cherry-picked into ExaDev/cddl.js#11 (released as 1.0.1), bumped here.

message-ref = { id: bstr, relation: tstr } gives room.send's optional refs array one open, typed relation ("reply", "forward", anything future) instead of a separate ad-hoc field per relation -- the same "open on purpose" precedent capability-scope.kind already set.

Nine new conformance vectors (7 frame, 2 token) round-trip through both the TS conformance test and Rust's conformance-check binary with zero Rust source changes -- confirms manage-command-params's Json catch-all arm and capability-scope's already-open kind field genuinely cover a brand new domain with no typed Rust arm needed, not merely assumed.

Test plan

• just spec && just conformance -- all 9 new vectors PASS in both TS and Rust
• just build && just test && just lint -- all green
• cargo test && cargo clippy --all-targets -- -D warnings && cargo fmt --check -- all green, zero Rust diff

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
🔒 Security Review	✅ Completed 2026-09-11T14:38:35.966987Z	ca35fd1	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.