feat(robsond): entry pipeline observability + debug endpoint

[ldamasio]

Summary

• Add entry-immediate observability with structured logging throughout entry pipeline
• Add debug endpoint for entry state inspection
• Refactor query engine and reconciliation worker for better traceability
• 18 files changed, ~780 insertions

Test plan

• Verify structured entry logs appear in robsond output
• Hit debug endpoint and confirm entry state is returned correctly
• Run cargo test — all existing tests pass

🤖 Generated with Claude Code

[ldamasio]

Closing without merge after value-risk review.

The initial split idea was to separate:

• observability-only changes;
• /debug/armed-positions and related debug state.

After a stricter value-risk review, we should not proceed with that split.

Findings:

• the PR adds no meaningful new metric or trace signal; the apparent metrics change is mostly churn/reformat around existing behavior;
• durable observability value collapses to a very small set of detector diagnostics;
• the /debug/armed-positions endpoint exposes operational trading state and must not land without a proper admin/access model;
• even the observability-only part includes info-level logging of sensitive operational fields such as symbol, side, account_id, entry mode, approval mode, entry price and stop loss;
• this is not aligned with the future Robson v4 / rbx-identity direction, where tenant/account scoping, admin/operator roles, redaction and auditability must be designed first.

Decision:

Close this PR entirely. Do not split and merge #82a by inertia.

Follow-up:

Create a new design issue for Robson administrative observability/debug access model, aligned with:

• rbx-identity;
• tenant_id / trading_account_id scoping;
• admin/operator roles;
• public vs admin vs debug API separation;
• redaction policy;
• audit logging;
• break-glass policy;
• safe metrics/logging rules;
• local/testnet-only debug gating.

Only if there is an active incident around immediate-entry diagnostics should we open a new tiny PR with just the minimal detector signals, without endpoint, debug state, broad lifecycle logs, docs churn or sensitive info-level logging.

Closing without merge.