feat(node): W1c — composition root, Node exposes 7 seams as Arc<dyn> handles (#1285)

[MichaelTaylor3d]

W1c — the #1285 composition-root milestone (Option A, locked by the orchestrator).

• Node stays ONE concrete struct implementing all 7 seam traits; the composition root exposes each seam as Arc<dyn SeamTrait> via 6 self-referential upcast constructors (as_chain_source/as_peer_network/as_rpc_dispatch/as_content_server/as_capsule_store/as_key_manager) — each Arc::clone(self) as Arc<dyn Trait> (same object, no new alloc). Consumers/FFI/tests hold Arc<dyn SeamTrait> (injectable fakes + per-seam swappable for W2-W5).
• dig-node-service AppState now holds content_server: Arc<dyn ContentServer>; loopback serve rewired through the seam handle (e2e demonstration).
• Behaviour-preserving: only the OUTER handle shape becomes Arc<dyn>; openrpc_drift_guard 7/7 green.
• Folded dig-urn-resolver 0.3.1 -> 0.4.0 (wire-compatible). Added SPEC §13 (seam architecture).
• MINOR bump: dig-node-core 0.14.0, binary 0.48.0 (additive new API + AppState field, §5.1).

Refs #1285. Inner seam de-tangle (multi-struct decomposition, cross-seam via trait handles, cycles broken by injection) deferred to #1357, sequenced after W2-W5 reshape the 4<->5 / 5<->6 edges.

[MichaelTaylor3d]

W1c correctness gate (fresh context, effort:high) — PASS on code.

Verified against PR head aaa087c (full diff = 7 files, +135/-10; git diff 3c72d75..aaa087c matches exactly — no hidden changes):

1. Upcasts are the SAME object, no new alloc — all 6 as_() are Arc::clone(self) as Arc<dyn Trait> (lib.rs), an Arc→Arc unsizing coercion. No Arc::new / Arc::new_cyclic, no state copy. Same object ⇒ same state ⇒ behaviour-preserving.
2. No behaviour change from the rewire — AppState.content_server = node.as_content_server() (server.rs build_state); the 3 loopback sites (serve_resource, serve_miss ×2) now dispatch through Arc to the SAME Node impl (impl ContentServer for Node, content_serve.rs:267). serve_content_plaintext + manifest_paths are trait methods with identical signatures — dynamic dispatch to the identical code.
3. openrpc_drift_guard — RPC surface untouched (rpc.rs/meta.rs/dispatch not in diff), so the 7/7 no-drift invariant cannot have drifted; the guard test runs in the CI 'Test + coverage' job.
4. No new tangle — diff introduces ONLY the 6 additive upcast ctors + the AppState field + 3 rewired sites. No new cross-seam reach; self.proxy/self.p2p_content/self.anchored_root_resolver are unchanged from base = exactly #1357's pre-existing scope.

Also: MINOR bump justified + both files agree (core 0.14.0, service 0.48.0, §5.1-additive: 6 new pub ctors + additive AppState field); dig-urn-resolver 0.3.1→0.4.0 present (wire-compatible); SPEC §13 accurate + references #1357; 0 unresolved review threads.

BEFORE MERGE (non-code, orchestrator): (a) PR title + body are STALE — title still reads '(DESIGN RECON, awaiting decider)' and the body is the recon note; the decision is locked (Option A) and this IS the implementation. The title becomes the squash-commit message, so update it to a clean conventional-commit summary that references #1285/#1357 (body should reference #1357). (b) CI: 'Test + coverage' and the native-package builds were still IN_PROGRESS at review time — merge only once all required checks are green.