fix(browser): recover from CDP DOM-node serialization crash in browser_console (#35385)

browser_console(expression="document.body") returned the cryptic CDP error
"Object reference chain is too long" instead of a usable result.

With returnByValue=true, Chrome deep-serializes the eval result; for a live
DOM Node/NodeList/Window that serialization overruns CDP's recursion guard
and fails the whole call with a protocol-level error (not a JS exception),
which _browser_eval surfaced raw.

- browser_supervisor.evaluate_runtime: on that specific error, retry once
  with returnByValue=false so Chrome returns the node's description string —
  the same graceful path already used for document.querySelector() results.
- browser_tool._browser_eval (CLI subprocess fallback): the subprocess can't
  retry, so convert the reference-chain error into actionable guidance
  (extract a primitive / use JSON.stringify) instead of leaking it raw.

No expression rewriting — normal evals (1+41 -> 42) are untouched.

tools/browser_supervisor.py

@@ -496,12 +496,12 @@ class CDPSupervisor:
         if not session_id:
             return {"ok": False, "error": "supervisor has no attached page session"}
 
-        async def _do_eval() -> Dict[str, Any]:
+        async def _do_eval(by_value: bool) -> Dict[str, Any]:
             return await self._cdp(
                 "Runtime.evaluate",
                 {
                     "expression": expression,
-                    "returnByValue": return_by_value,
+                    "returnByValue": by_value,
                     "awaitPromise": await_promise,
                     # userGesture matters for things like clipboard / fullscreen
                     # APIs that require a user-activation context.
@@ -511,14 +511,32 @@ class CDPSupervisor:
                 timeout=timeout,
             )
 
-        try:
-            from agent.async_utils import safe_schedule_threadsafe
-            fut = safe_schedule_threadsafe(_do_eval(), loop)
+        from agent.async_utils import safe_schedule_threadsafe
+
+        def _run_eval(by_value: bool) -> Dict[str, Any]:
+            fut = safe_schedule_threadsafe(_do_eval(by_value), loop)
             if fut is None:
-                return {"ok": False, "error": "Browser supervisor loop unavailable"}
-            response = fut.result(timeout=timeout + 1)
+                raise RuntimeError("Browser supervisor loop unavailable")
+            return fut.result(timeout=timeout + 1)
+
+        try:
+            response = _run_eval(return_by_value)
         except Exception as exc:
-            return {"ok": False, "error": f"{type(exc).__name__}: {exc}"}
+            # ``returnByValue=True`` asks Chrome to deep-serialize the result.
+            # For live DOM nodes / NodeLists / Window that serialization can
+            # blow past CDP's recursion guard and fail the whole call with
+            # ``Object reference chain is too long`` (a protocol-level error,
+            # not a JS exception).  Retry once with ``returnByValue=False`` so
+            # Chrome returns the object's description string instead — the same
+            # graceful degradation path used for ``document.querySelector(...)``
+            # results — rather than crashing the eval.
+            if return_by_value and "reference chain is too long" in str(exc).lower():
+                try:
+                    response = _run_eval(False)
+                except Exception as exc2:
+                    return {"ok": False, "error": f"{type(exc2).__name__}: {exc2}"}
+            else:
+                return {"ok": False, "error": f"{type(exc).__name__}: {exc}"}
 
         # Runtime.evaluate response shape:
         #   {"id": N, "result": {"result": {"type": "...", "value": ..., ...},