Fail fast when peer has no ring location

[sanity]

Nacho noted in PR #2064 that we should not route operations before a peer has finished joining the network, and more generally that we keep conflating network subscriptions with local client subscriptions.

Current behavior:

• PUT/SUBSCRIBE/etc. traffic still flows during the short window where connection_manager.own_location() returns PeerKeyLocation { location: None }, so the router fallback exists to avoid dropping those entries.
• Local client subscriptions share the same routing path as network seeding, leading to repeated complications in update delivery.

Follow-up plan:

1. Decide the correct behavior when the local node lacks a ring location (likely fail fast with a clear “peer not joined” error).
2. If we fail fast, propagate that error from PUT/SUBSCRIBE/GET/etc. and teach callers/tests (including client subscription handling) to retry after join completes.
3. Revisit the architecture that conflates network subscriptions and local client subscriptions—consider a separate flow (e.g., internal notifications) so network routing isn’t used for purely local listeners.
4. Once the above is in place, remove the router fallback (peer_distance_or_default) so routing only considers peers with valid locations.

Context:

• feat: harden subscription routing #2064 (comment)
• feat: harden subscription routing #2064 (comment)
• feat: harden subscription routing #2064 (comment)

[github-actions]

🤖 Auto-labeled

Applied labels:

• T-bug (92% confidence)
• P-medium (82% confidence)
• E-medium (78% confidence)
• A-networking (96% confidence)
• S-needs-design (85% confidence)

Reasoning: The report describes incorrect behavior (operations are routed before the node has a ring location) and proposes changing behavior to fail fast and propagate errors, which aligns with a bug fix rather than new functionality. The impact is moderate (should be fixed but not clearly release-blocking), so P-medium is appropriate. Implementing requires touching multiple entry points and tests and removing Router fallback, indicating moderate effort. The issue concerns ring location and routing, so A-networking applies. The first follow-up is a decision about expected behavior, so S-needs-design is warranted.

Previous labels: none

If these labels are incorrect, please update them. This helps improve the auto-labeling system.

[sanity]

We spun out the broader architectural clean-up into #2075 (decoupling local vs network subscriptions) so this issue can stay focused on the “peer not joined” fail-fast semantics and removing the routing fallback once that’s in place. The two issues are related but #2075 tracks the wider refactor Nacho raised in #2064.