ws: test gaps from PR #136 review — Phase 3/5 multi-node, TCP RST, --mcp fallback

[TickTockBent]

Follow-up from PR #136 review. Tracks test coverage gaps not blocking merge but worth filling before Phase 7 burn-in.

Gaps

1. Phase 3 cross-enclave-peer fanout — TestWSPutStoresLocallyAndAcks covers the local-store side of a relayed PUT but not the criterion "Transient PUT shows up in all enclave-peer stores within gossip latency." Needs a 2-substrate test fixture.

2. Phase 5 receive path multi-node variants — "two-substrate, transient on each" path; "transient A writes, transient B sees it" path. Both stop short of being tested end-to-end.

3. TCP-RST acceptance criterion — issue mcp: port WS transport + tree manager to Go (Option A — restore NAT-traversal / transient participation) #135 calls out "Forced TCP RST surfaces a clean disconnect within 30s." Currently only graceful Close() is tested.

4. --mcp multi-seed fallback — cmd/repram/main.go:281-310 loops through peers and gives up after all fail. No automated test; would need an observability hook (returned error / metric / log capture).

5. Heartbeat at production cadence (30s) — unit tests use 30ms; the 5-min idle acceptance is implicit. Phase 7 burn-in covers this.

Related: #131 (Phase 4 multi-node quorum_status="pending" test).

Acceptance

• At least items 1, 2, 3 covered before Phase 7 burn-in begins
• Items 4-5 covered or explicitly deferred

// ticktockbent