probing: add cancel command for stuck measurements

[matthieugouel]

Adds nxthdr probing cancel <id> to resolve a stuck/in-progress measurement — pairs with the gateway endpoint in saimiris-gateway#47 (POST /api/measurement/{id}/cancel). (roadmap nxthdr/roadmap#38)

Changes

• probing cancel <id> → POSTs the cancel endpoint, reports cancelled / agents_cancelled (idempotent — a no-op with a message if already terminal).
• Surfaces the new cancelled state everywhere: measurements (list) and measurement-status now show status cancelled (and per-agent cancelled), via a shared measurement_label(cancelled, complete) helper. New measurement_cancelled fields are #[serde(default)], so the CLI still works against a gateway that predates the cancel deploy.

Testing

cargo build + cargo clippy --all-targets clean. End-to-end test against prod once gateway #47 is deployed (cancel the real stuck cf78cc3d… measurement).

Order: merge + deploy gateway #47 before this lands, so cancel hits a gateway that has the endpoint.

🤖 Generated with Claude Code

[matthieugouel]

Claude finished @matthieugouel's task in 1m 32s —— View job

---

PR Review

• Gather context and read changed files
• Analyze diff for logic bugs, security issues, and CLAUDE.md violations
• Post review findings

---

Two findings worth addressing before merge:

1. message field silently dropped in JSON/CSV mode

In cancel() (src/probing.rs:396-400), resp.message is only surfaced via output::success/output::info, which are text-only helpers. In -o json or -o csv mode the message is silently lost — a machine consumer only sees cancelled: false with no reason why.

CLAUDE.md says data must be routed through output.rs so -o json/-o csv keep working. Adding message to the kv block fixes this:

output::kv(&[
    ("id", id),
    ("cancelled", &resp.cancelled.to_string()),
    ("agents_cancelled", &resp.agents_cancelled.to_string()),
    ("message", &resp.message),   // ← add this
]);

Fix this →

2. Per-agent table header "complete" is now stale

In measurement_status() (src/probing.rs:371), the agent status column now renders three distinct values — "cancelled", "yes", or "no" — but the table header still says "complete":

output::table(&["agent", "probes sent/expected", "complete"], &rows);
//                                                ^^^^^^^^^^
// now shows "cancelled" / "yes" / "no" — header is misleading

This affects text, JSON, and CSV output (the header becomes a JSON key and a CSV column name). A header like "status" would be accurate across all three values.

Fix this →

---

The rest of the implementation looks solid: #[serde(default)] backward-compatibility on the new fields is correct, the measurement_label precedence (cancelled > complete > in-progress) is right, and cancel correctly uses new_saimiris(). No security issues.