CDP telemetry fixes (#251)

## Overview

### Bug

When chromium restarted mid-session (`PATCH /chromium/flags`,
`/configure`, etc., or a crash + supervisord autorestart), the CDP proxy
emitted `cdp_disconnect.reason = "client_close"` instead of
`"upstream_changed"`. Live repro confirmed: ~1.6s after the WS opened,
reason was `client_close` even though chromium clearly went away.

### Root cause

A goroutine watched `mgr.Subscribe()` for the new chromium URL and
stamped `reasonOverride = upstream_changed` before triggering pump
teardown. But the upstream's TCP EOF (chromium dying) reaches the pump
*before* the new "DevTools listening on" log line is emitted (~5–8 s
gap). So `cleanup` always ran first, saw `reasonOverride == nil`, and
fell through to `client_close`.

### Fix

- `wsproxy.Pump`'s `onClose` callback now receives a `PumpExitCause`
(`client` / `upstream` / `context`).
- Deleted the URL-watcher goroutine + `reasonOverride` in
`devtoolsproxy`. Reason resolution moved into cleanup as a small helper,
`resolveDisconnectReason`, which:
- returns `client_close` / `context_cancelled` immediately from the
cause,
- on upstream cause, checks `mgr.Current()` and otherwise waits up to
`restartConfirmWait` (10 s) on `urlCh` for a new URL — new URL ⇒
`upstream_changed`, timeout ⇒ `upstream_error`.

### Tests

- `TestResolveDisconnectReason` — 6 table cases covering all four
reasons (sub-100 ms).
- `TestWebSocketProxyHandler_EmitsUpstreamChangedOnMidStreamRestart` —
integration through the handler with `restartConfirmWait` shortened.
- Existing dial-time `upstream_error` + client_close tests continue to
pass.

### Verification

Repro'd live against the headful image. Before fix:
`reason=client_close, duration_ms=1620`. After fix:
`reason=upstream_changed, duration_ms=10388`.


<!-- CURSOR_SUMMARY -->
---

> [!NOTE]
> **Medium Risk**
> Changes CDP WebSocket teardown and telemetry classification on
Chromium restarts; well-tested but affects automation clients that rely
on disconnect reasons and session timing.
> 
> **Overview**
> Fixes **CDP disconnect telemetry** when Chromium restarts mid-session:
`cdp_disconnect` was often emitted as `client_close` because upstream
EOF arrived before the new DevTools URL was published.
> 
> **`wsproxy.Pump`** now passes a **`PumpExitCause`** (`client` /
`upstream` / `context`) into cleanup. **`devtoolsproxy`** drops the
URL-watcher `reasonOverride` path; on upstream exit it
**`resolveDisconnectReason`** polls `UpstreamManager.Current()` for up
to **`restartConfirmWait`** (10s) to choose **`upstream_changed`** vs
**`upstream_error`**. When the upstream URL changes while a session is
still open, the watcher **closes the upstream WebSocket** so the pump
exits with the right cause. **`cdp_disconnect`** uses a pinned
**`disconnectedAt`** for **`duration_ms`** and event timestamp so
metrics are not inflated by the restart poll.
> 
> New unit/integration tests cover reason resolution, mid-stream
restart, and proactive kick on URL change. The rest of the diff is
mostly **import ordering**, **gofmt**, and comment/whitespace cleanup
across API, events, policy, and e2e helpers.
> 
> <sup>Reviewed by [Cursor Bugbot](https://cursor.com/bugbot) for commit
f43d049. Bugbot is set up for automated
code reviews on this repo. Configure
[here](https://www.cursor.com/dashboard/bugbot).</sup>
<!-- /CURSOR_SUMMARY -->

---------

Co-authored-by: Sayan- <1415138+Sayan-@users.noreply.github.com>

Author: Sayan-

server/cmd/api/api/api.go

@@ -12,14 +12,14 @@ import (
 
 	"github.com/kernel/kernel-images/server/lib/cdpmonitor"
 	"github.com/kernel/kernel-images/server/lib/devtoolsproxy"
-	"github.com/kernel/kernel-images/server/lib/telemetry"
 	"github.com/kernel/kernel-images/server/lib/events"
 	"github.com/kernel/kernel-images/server/lib/logger"
 	"github.com/kernel/kernel-images/server/lib/nekoclient"
 	oapi "github.com/kernel/kernel-images/server/lib/oapi"
 	"github.com/kernel/kernel-images/server/lib/policy"
 	"github.com/kernel/kernel-images/server/lib/recorder"
 	"github.com/kernel/kernel-images/server/lib/scaletozero"
+	"github.com/kernel/kernel-images/server/lib/telemetry"
 )
 
 type cdpMonitorController interface {
@@ -99,7 +99,7 @@ func New(
 	stz scaletozero.PinnedController,
 	nekoAuthClient *nekoclient.AuthClient,
 	telemetrySession *telemetry.TelemetrySession,
-	eventStream      *events.EventStream,
+	eventStream *events.EventStream,
 	displayNum int,
 ) (*ApiService, error) {
 	switch {

server/cmd/api/api/api_test.go

@@ -12,12 +12,12 @@ import (
 	"log/slog"
 
 	"github.com/kernel/kernel-images/server/lib/devtoolsproxy"
-	"github.com/kernel/kernel-images/server/lib/telemetry"
 	"github.com/kernel/kernel-images/server/lib/events"
 	"github.com/kernel/kernel-images/server/lib/nekoclient"
 	oapi "github.com/kernel/kernel-images/server/lib/oapi"
 	"github.com/kernel/kernel-images/server/lib/recorder"
 	"github.com/kernel/kernel-images/server/lib/scaletozero"
+	"github.com/kernel/kernel-images/server/lib/telemetry"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 )

server/cmd/api/api/computer_test.go

@@ -262,7 +262,7 @@ func TestGaussianDelay_WelfordVelocityVariance(t *testing.T) {
 	for i := range distances {
 		t_norm := float64(i) / float64(steps)
 		base := 5.0 + 15.0*math.Sin(t_norm*math.Pi) // 5-20px, peaked in middle
-		distances[i] = base + rng.Float64()*3.0       // small random variation
+		distances[i] = base + rng.Float64()*3.0     // small random variation
 	}
 
 	// Gaussian delays → velocity variance
@@ -302,27 +302,27 @@ func TestClampPoints(t *testing.T) {
 		expected [][2]int
 	}{
 		{
-			name:     "no clamping needed",
-			points:   [][2]int{{10, 20}, {50, 50}, {100, 80}},
-			w:        200, h: 200,
+			name:   "no clamping needed",
+			points: [][2]int{{10, 20}, {50, 50}, {100, 80}},
+			w:      200, h: 200,
 			expected: [][2]int{{10, 20}, {50, 50}, {100, 80}},
 		},
 		{
-			name:     "clamp negative x and y",
-			points:   [][2]int{{-10, -20}, {50, 50}},
-			w:        200, h: 200,
+			name:   "clamp negative x and y",
+			points: [][2]int{{-10, -20}, {50, 50}},
+			w:      200, h: 200,
 			expected: [][2]int{{0, 0}, {50, 50}},
 		},
 		{
-			name:     "clamp exceeding screen bounds",
-			points:   [][2]int{{50, 50}, {250, 300}},
-			w:        200, h: 200,
+			name:   "clamp exceeding screen bounds",
+			points: [][2]int{{50, 50}, {250, 300}},
+			w:      200, h: 200,
 			expected: [][2]int{{50, 50}, {199, 199}},
 		},
 		{
-			name:     "clamp both directions",
-			points:   [][2]int{{-5, 250}, {300, -10}, {100, 100}},
-			w:        200, h: 200,
+			name:   "clamp both directions",
+			points: [][2]int{{-5, 250}, {300, -10}, {100, 100}},
+			w:      200, h: 200,
 			expected: [][2]int{{0, 199}, {199, 0}, {100, 100}},
 		},
 	}

server/cmd/api/api/display.go

@@ -12,11 +12,11 @@ import (
 	"strings"
 	"time"
 
-	nekooapi "github.com/m1k1o/neko/server/lib/oapi"
 	"github.com/kernel/kernel-images/server/lib/cdpclient"
 	"github.com/kernel/kernel-images/server/lib/logger"
 	oapi "github.com/kernel/kernel-images/server/lib/oapi"
 	"github.com/kernel/kernel-images/server/lib/recorder"
+	nekooapi "github.com/m1k1o/neko/server/lib/oapi"
 )
 
 // PatchDisplay updates the display configuration. When require_idle
@@ -655,4 +655,3 @@ func (s *ApiService) setResolutionViaNeko(ctx context.Context, width, height, re
 	log.Info("successfully changed resolution via Neko API", "width", width, "height", height, "refresh_rate", refreshRate)
 	return nil
 }
-

server/cmd/api/api/fs.go

@@ -16,11 +16,11 @@ import (
 	"os/user"
 
 	"github.com/fsnotify/fsnotify"
-	"github.com/nrednav/cuid2"
 	"github.com/kernel/kernel-images/server/lib/logger"
 	oapi "github.com/kernel/kernel-images/server/lib/oapi"
 	"github.com/kernel/kernel-images/server/lib/ziputil"
 	"github.com/kernel/kernel-images/server/lib/zstdutil"
+	"github.com/nrednav/cuid2"
 )
 
 // fsWatch represents an in-memory directory watch.

server/cmd/api/api/process.go

@@ -23,10 +23,10 @@ import (
 	"github.com/coder/websocket"
 	"github.com/creack/pty"
 	"github.com/google/uuid"
-	openapi_types "github.com/oapi-codegen/runtime/types"
 	"github.com/kernel/kernel-images/server/lib/logger"
 	oapi "github.com/kernel/kernel-images/server/lib/oapi"
 	"github.com/kernel/kernel-images/server/lib/ptyio"
+	openapi_types "github.com/oapi-codegen/runtime/types"
 )
 
 type processHandle struct {

server/cmd/api/api/process_test.go

@@ -13,9 +13,9 @@ import (
 	"time"
 
 	"github.com/google/uuid"
-	openapi_types "github.com/oapi-codegen/runtime/types"
 	oapi "github.com/kernel/kernel-images/server/lib/oapi"
 	"github.com/kernel/kernel-images/server/lib/scaletozero"
+	openapi_types "github.com/oapi-codegen/runtime/types"
 	"github.com/stretchr/testify/require"
 )
 

server/cmd/chromium-launcher/main.go

@@ -135,7 +135,6 @@ func main() {
 	}
 }
 
-
 // execLookPath helps satisfy syscall.Exec's requirement to pass an absolute path.
 func execLookPath(file string) (string, error) {
 	if strings.ContainsRune(file, os.PathSeparator) {

server/cmd/chromium-launcher/main_test.go

@@ -34,4 +34,3 @@ func TestExecLookPath(t *testing.T) {
 		t.Fatalf("execLookPath PATH search failed: p=%q err=%v", p, err)
 	}
 }
-

server/cmd/shell/main.go

@@ -17,8 +17,8 @@ import (
 	"time"
 
 	"github.com/google/uuid"
-	openapi_types "github.com/oapi-codegen/runtime/types"
 	oapi "github.com/kernel/kernel-images/server/lib/oapi"
+	openapi_types "github.com/oapi-codegen/runtime/types"
 	"golang.org/x/term"
 )