Refactor gpu-remoting workload sender scripts

[Reggie1314129]

Summary

This PR refactors duplicated workload sender logic in gpu-remoting by extracting the shared command-building and trace-replay flow into a common module.

Changes

• add a shared workload_sender.py module
• move common workload command construction into the shared module
• move shared trace replay logic into the shared module
• simplify client_sender.py into a thin entrypoint for the default gpu-remoting environment
• simplify scheduler/cluster_client_sender.py into a thin entrypoint for the scheduler/cluster environment
• preserve environment-specific differences such as dataset paths and trace-specific overrides

Why

Both sender scripts previously duplicated nearly the same implementation, which made them harder to maintain and easy to drift apart. This change keeps existing behavior while reducing duplication and making future sender-side updates easier.

Validation

• verified updated Python files with python3 -m py_compile