fix: record shutdown interruptions and keep scan cancellation requests

[crypt0rr]

Summary

Restarts no longer look like cancellations (#1227)

• A scan that stops because EdgeWatch stopped (shutdown, upgrade, or loss of the daemon lease) keeps the canceled status. Its error now reads "scan interrupted because EdgeWatch stopped"; a resumable attempt adds "; progress was saved".
• It emits a new scan-interrupted event. Activity shows it as Scan interrupted, and it is never queued for notification destinations (model.EventDelivered, checked in queueEventsTx). A restart therefore no longer sends one "Scan canceled" alert per destination. The job-silence alert still reports a daemon that keeps stopping.
• How a scan is classified:
  • Interrupted: the run context ended and nobody asked to cancel the scan.
  • Canceled, as today: someone asked to cancel the scan, either through Cancel scan or by pausing the business unit. This includes a cancel request that arrives during a shutdown.

Cancellation state survives progress updates (#1228)

• Active scans now report cancel_requested: true once a cancel is requested. phase keeps following the scanner, which reports one last progress update when its process stops, so finalizing stays visible.
• The job page and the Overview's active-scan row both show Cancellation requested instead of the Cancel scan button while the flag is set. Before this change the Overview row re-enabled the button as soon as the cancel request returned.
• Once finalization has started, POST /scans/{id}/cancel returns 409 scan_finalizing, because cancelling could no longer change the result.

model.Fingerprint (#1237)

• It sorts a copy of the CPEs instead of the caller's slice. The defensive copy in acceptedServiceForEvidence is no longer needed and has been removed. The fingerprint values are unchanged.

Compatibility

• New fields and values:
  • cancel_requested on active scans;
  • a scan-interrupted event type in Activity and the live stream;
  • a 409 scan_finalizing from the cancel route.
• Behaviour change: a scan interrupted by shutdown no longer queues a scan-canceled notification.
• Unchanged: the schema, scan status values, and the user-cancel alert (still pinned by TestManagedTerminalOutcomesQueueNotifications).

Validation

• gofmt, go vet ./...
• go test -race for internal/app, internal/engine, internal/model and internal/web, plus the outbox, runtime and incident tests in internal/store.
• npm run lint
• Vitest for main, Activity, Dashboard and JobDetail.
• npm --prefix docs run build

New tests:

• internal/app/scan_interruption_test.go:
  • stopping the run context during a direct scan and during a resumable scan stores the interruption text, returns scan-interrupted, writes it to Activity, and queues nothing in the outbox;
  • cancel_requested survives a progress update;
  • a cancel after finalization begins returns ErrScanFinalizing;
  • classification edge cases (timeout, request during shutdown, no run).
• Engine: cases for interrupted scans, direct and paused.
• Model: EventDelivered, and Fingerprint leaves the caller's slice unchanged.
• Console: the Overview and job page keep the pill while the phase is scanning or finalizing; Activity labels the event; the stream invalidates for it.

Docs: the notifications guide covers interrupted scans, and the scanning guide covers the cancel flow.

Fixes #1227
Fixes #1228
Fixes #1237

[cloudflare-workers-and-pages]

Deploying edgewatch with    Cloudflare Pages

Latest commit:	67f0788
Status:	✅  Deploy successful!
Preview URL:	https://e5968c1b.edgewatch-cpd.pages.dev
Branch Preview URL:	https://fix-scan-cancellation-causes.edgewatch-cpd.pages.dev

View logs