fix: linkspan socket in sticky /tmp, not a shared csbridge/ dir

[yasithdev]

Summary

linkspan's unix socket lived at /tmp/csbridge/<id>.sock, and the SLURM job script ran mkdir -p /tmp/csbridge. On a shared compute node, the first user to launch a session there ends up owning /tmp/csbridge at mode 0755. Every subsequent different user's mkdir -p is a silent no-op (the dir already exists), and linkspan's bind() on the socket fails:

failed to listen on unix socket /tmp/csbridge/<id>.sock: bind: permission denied

Observed live on the Nexus cluster: /tmp/csbridge on node c2 was drwxr-xr-x 2000043 2000043, so nexus-yjayawardana (uid 2000047, only in the group — no group write) couldn't create his socket and every session on that node failed with exit 1:0.

Fix

Bind the socket at /tmp/csbridge-<id>.sock — directly in the sticky, world-writable /tmp — instead of inside a shared csbridge/ subdirectory. Each user creates and owns their own socket; the sticky bit stops others deleting it; there is no shared directory to clash over. The mkdir -p line is gone (nothing to create). The socket name already carries the unique session id, so slurmSupport's srun --overlap curl --unix-socket … channel follows automatically, and linkspan needs no change (it binds whatever --socket it's handed and clears its own stale path via os.Remove).

Test plan

• npm run check-types — clean
• npm test — 108/108 pass (updated the buildSlurmScript assertion to the new path)
• npm run package — builds
• Generated script verified: --socket /tmp/csbridge-<id>.sock, no mkdir, no /tmp/csbridge/ dir reference

🤖 Generated with Claude Code