Fix race condition in new_session() by avoiding list-sessions query

[neubig]

Fixes: #624

Previously, new_session() would:

1. Run 'tmux new-session -P -F#{session_id}' to create session
2. Immediately run 'tmux list-sessions' to fetch full session data

This created a race condition in some environments where list-sessions might not see the newly created session yet, causing TmuxObjectDoesNotExist errors.

The fix:

• Expand the -F format string to include all Obj fields (except 'server')
• Parse the output directly into a Session object
• Eliminate the separate list-sessions query entirely

This is more efficient (one fewer subprocess call) and eliminates the race condition by making session creation atomic.

I tested this using the script in the original issue and it works in all three cases:

Environment	Result
Python 3.12 + PyInstaller + Docker	PASS ✓
Python 3.13 + PyInstaller + Docker	PASS ✓
Python 3.13 + PyInstaller + Native	PASS ✓

[CLAassistant]

All committers have signed the CLA.

[tony]

@neubig Thank you. Can you sign the CLA?

Is there a CI or full traceback for this? Any tmux version you're trying with? Thanks for the details in #624.

[neubig]

Thanks @tony! Signed the CLA.

I checked and the version is tmux 3.5a.

And in terms of CI where this failed, the reason why we were able to identify this is because this CI run failed: https://github.com/OpenHands/software-agent-sdk/actions/runs/21638051841/job/62369267088

We were able to trace it back to the upgrade from Python 3.12 to 3.13, and after reverting to 3.12, the consistent failures due to the issue I cited here disappeared (although this particular run still has a few failures): https://github.com/OpenHands/software-agent-sdk/actions/runs/21710957108