feat(telemetry): step seconds and a gate counter for the stage row #4066

Problem

Stage has one metric, devops_cli_roadmap_stage_total, with the label result only (stage.py, emit_stage_results). No metric shows how long the steps of a stage round take. No counter shows which gate refused a plan. check_proposal joins the problems of all gates, and the gate name is lost.

Expected

Add one histogram devops_cli_roadmap_step_seconds with the labels repo, row, step and result. The step values come from StepName (roadmap/failure.py). The row name is passed as an argument, as tool_metrics.start_run(row=...) takes it. So no shared scope is needed.

In this issue row is stage. The issue #4067 reuses the histogram for the implement row.

StepName members for the stage:

• plan and apply exist from the issue feat(roadmap): record the failure class and step of a stage item #4056.
• This issue adds select, gate and review.

What each step times:

• select: read_stage_selection.
• plan: the model calls of the planner for one item.
• gate: the calls of the gates for one item.
• review: the model review of the verifier.
• apply: the write of one item.

result has two values: ok and error.

Series count: 1 repo x 1 row x 5 steps x 2 results = 10 series, times the histogram buckets.

Add the counter devops_cli_roadmap_stage_gate_refusals_total with the labels repo and gate. The planner calls check_gates where it calls check_proposal, and adds 1 for each gate that found a problem. The value of gate is a ProposalGate value. Series count: 14 gates x 1 repo = 14.

Document both series in docs/TELEMETRY.md. State the series counts there.

No live model run (AGENTS.md rule 13).

Acceptance criteria

• uv run pytest tests/telemetry/test_roadmap_stage_steps.py -k step_label - a stage round emits a step sample for select, plan, gate, review and apply.
• uv run pytest tests/telemetry/test_roadmap_stage_steps.py -k gate_refusal - a refused plan increments the counter of its gate name, and only that gate.
• uv run pytest tests/telemetry/test_roadmap_stage_steps.py -k step_names - the step values are members of StepName. No other value appears.
• grep -q 'devops_cli_roadmap_step_seconds' docs/TELEMETRY.md && grep -q 'devops_cli_roadmap_stage_gate_refusals_total' docs/TELEMETRY.md - the documentation lists both series (it prints nothing today).

Every command above fails on the unchanged code. The pull request cites the fail-to-pass run (devops ci fail-to-pass --base <ref>).

Feasibility

I read these files at origin/main: stage.py (emit_stage_results, _settle, read_stage_selection), stage_change.py (check_gates, check_proposal, ProposalGate), roadmap/tool_metrics.py, telemetry/instruments.py and docs/TELEMETRY.md. I checked that check_proposal calls check_gates and joins its lists. I did not run the code. The change makes no GitHub request.

Existing tool considered

telemetry/instruments.py (emit), tool_metrics.py as the pattern, and the OpenTelemetry instruments that exist. No new library.

Cost

Build: about 50k tokens and 2.5 h. Keep: about 3k tokens per quarter. Fix: about 10k tokens and 0.5 h.

Related

Part of #3027. #3069, #3650, #3655, #3162, #3566. Order: after the issues #4056 and #4064. All of them change stage.py or the telemetry files.