fix(web): keep the action status fetch alive when a client disconnects

The status fetch used the request context. A client that dropped the
connection, such as a readiness probe with a one second timeout,
cancelled the fetch. The failure was then cached and every later
request showed no action status. The fetch now ignores the caller
cancellation and keeps its own timeout.

internal/web/actions.go

@@ -158,7 +158,8 @@ func (a *Actions) statusRows(ctx context.Context, rt *actionRuntime) (map[alertR
 	if !rt.fetched.IsZero() && a.now().Sub(rt.fetched) < statusCacheTTL {
 		return rt.rows, rt.err
 	}
-	rows, err := a.fetchStatus(ctx, rt.cfg)
+	// A client that disconnects mid-request must not fail the fetch that every later request reuses.
+	rows, err := a.fetchStatus(context.WithoutCancel(ctx), rt.cfg)
 	if err != nil {
 		a.log.Warn("action status unavailable", "action", rt.cfg.ID, "err", err)
 	}

internal/web/actions_test.go

@@ -2,6 +2,7 @@ package web
 
 import (
 	"bytes"
+	"context"
 	"encoding/json"
 	"io"
 	"log/slog"
@@ -407,3 +408,15 @@ func TestRedirectDoesNotForwardToken(t *testing.T) {
 		t.Fatal("bearer token forwarded across a redirect")
 	}
 }
+
+func TestStatusFetchSurvivesCancelledRequest(t *testing.T) {
+	fh := newFakeHook(t)
+	fh.statBody = `[{"repo":"o/r","url":"` + alertURL + `","state":"running"}]`
+	h := newActionServer(t, nil, fh.srv.URL)
+	ctx, cancel := context.WithCancel(context.Background())
+	cancel()
+	rows, err := h.actions.statusRows(ctx, h.actions.list[0])
+	if err != nil || len(rows) != 1 {
+		t.Fatalf("rows=%d err=%v, want the fetch to complete despite the cancelled request", len(rows), err)
+	}
+}