[feature] status/inspect: report waiting_for per node and document R1 acceptance gating #52

Split from #50 (item 2a).

Problem

From status and inspect, an operator can't tell why a node is waiting. In the holoself sprint run the operator expected only Tier3 nodes to stop for accept. When a Tier2 node stopped at AwaitingReview, they had to read the source to find out why.

In the current R1 subset (ADR 0016) every node stops for human acceptance, and the tier only decides routing. ADR 0007 requires human acceptance only at Tier 3, so expecting gating by tier is reasonable, and the difference isn't documented anywhere a user would see it.

Proposal

1. Add a waiting_for field to each node in status --json and inspect --json, with one of:

   • acceptance (state AwaitingReview)
   • dependency (with the blocking node ids)
   • plan_acceptance
   • retry_budget_exhausted
   • null when the node isn't blocked

   The text output shows the same reason.

2. docs/start.md, plus the meshloop-review-plan and meshloop-accept skills, say plainly: in R1, every node stops for meshloop accept, and the tier only chooses the harness and model.

3. review-plan output lists which nodes will need accept, which today is all of them.

Changing which nodes stop is proposed separately in #53. This issue only reports the current behaviour accurately.

Acceptance criteria

• Test: a node in AwaitingReview has waiting_for: "acceptance". A node waiting on another node has waiting_for: "dependency" and lists that node's id.
• docs/start.md and the two skills state the R1 acceptance rule.

🤖 Generated with Claude Code