Recording session randomly stops in LeLab with no error shown in GUI

When starting a training/recording session, LeLab sometimes stops recording
partway through with no error message in the GUI. The backend log shows the
camera and arm disconnecting and the session being marked as failed, but no
underlying exception/traceback is printed or surfaced to the frontend. This is
happening on the most recent version of LeLab, on Windows 11.

Steps to reproduce

Start a recording session in LeLab. Specify episode number (i.e: 50).
Recording runs fine, but intermittently stops early (i.e: episode 13.
GUI shows no error. Status just goes to "failed" silently.
Logs:

INFO 2026-07-17 11:49:33 a_opencv.py:601 OpenCVCamera(0) disconnected.
INFO 2026-07-17 11:49:33 a_opencv.py:601 OpenCVCamera(1) disconnected.
INFO 2026-07-17 11:49:33 follower.py:240 so-arm101 SOFollower disconnected.
INFO 2026-07-17 11:49:33 o_leader.py:163 so-arm101 SOLeader disconnected.
ERROR 2026-07-17 11:49:33 b\record.py:312 Recording session failed
Svt[info]: Only a single picture was passed in. Consider setting avif=1 for improved efficiency and reduced memory usage.
Svt[info]: Only a single picture was passed in. Consider setting avif=1 for improved efficiency and reduced memory usage.
INFO 2026-07-17 11:49:34 b\record.py:327 Recording session ended
INFO 2026-07-17 11:49:34 b\record.py:404 📡 RECORDING STATUS REQUEST: Session failed with error - frontend should stop polling
📡 STATUS CHANGE: Frontend is still polling after session error - should stop now

Start with b\record.py around lines 312, 327, and 404, then inspect the camera and arm disconnect entries in a_opencv.py, follower.py, and o_leader.py. Trace the frontend status polling mentioned in the log and reproduce the intermittent failure on Windows 11. Done means the recording failure exposes its underlying exception or traceback and the GUI presents the error instead of failing silently.

Written by the indexing model from the issue text.