[codex] Use atomic write in _update_upload_status to prevent file corruption

[rahulrao85]

Summary

Replace the seek(0) → json.dump → truncate() in-place rewrite pattern in _update_upload_status() with a temp-file + os.replace() atomic write pattern.

Problem

The previous implementation used:

f.seek(0)
json.dump(data, f, indent=2)
f.truncate()

If json.dump() raised an exception mid-write (e.g. serialization error for a non-serializable value), the file was left with partially-overwritten data at a wrong truncation position — corrupting the session JSON file permanently. Both the org and personal uploaders use this same function, so a crash in either process would corrupt the shared session file.

Fix

1. Read the session file under the exclusive lock (unchanged)
2. Write the full JSON to a .tmp sibling file using json.dump
3. Atomically swap with os.replace(tmp_path, session_file)

If json.dump crashes, the .tmp file is cleaned up and the original session_file remains untouched. os.replace() is atomic on POSIX and modern Windows (NTFS atomic rename), so readers always see a consistent file.

Validation

• Same locking semantics preserved (exclusive lock protects the read)
• os.replace is used instead of shutil.move for cross-filesystem safety (ensures atomic rename when source and dest are on the same filesystem)
• Temp file cleanup in except block prevents orphaned .tmp files