Refine known_failures cache update handling (#3744)

## Changes
- separate downloading of known_failures.txt from writing it to disk
- maintain atomic replacement of the cached known_failures file

------
https://chatgpt.com/codex/tasks/task_e_68e7c56a98d88325aae2335e5f37e6b0

Original PR: denik#4

Author: denik

tools/gh_report.py

@@ -130,12 +130,17 @@ def download_known_failures():
         if file_age < 300:  # 5 minutes
             return
 
+    temp_path = known_failures_path.with_name(f"{known_failures_path.name}.{os.getpid()}")
+
     try:
         with urllib.request.urlopen(KNOWN_FAILURES_URL) as response:
             content = response.read().decode("utf-8")
-        known_failures_path.write_text(content)
     except Exception as e:
         print(f"Failed to download known_failures.txt: {e}", file=sys.stderr)
+        return
+
+    temp_path.write_text(content)
+    temp_path.replace(known_failures_path)
 
 
 def download_run_id(run_id, repo, rm):