Run long-lived background commands in a Daytona sandbox using process sessions and stream their stdout/stderr in real time

Documented steps

1. Create a persistent session first with `create_session(session_id)` — sessions are what allow a command to outlive a single request.
2. Launch the command with `execute_session_command(session_id, SessionExecuteRequest(...))`, setting `run_async=True` (Python) / `runAsync: true` (TypeScript) so the call returns immediately instead of blocking.
3. Capture the returned command id (`cmdId`); it is required for polling status, sending input, and streaming logs.
4. Stream output live with Python `sandbox.process.get_session_command_logs_async(session_id, cmd_id, stdout_callback, stderr_callback)` or TypeScript `sandbox.process.getSessionCommandLogs(sessionId, cmdId, stdoutCallback, stderrCallback)`.
5. For a point-in-time snapshot instead of a stream, call `get_session_command_logs(session_id, cmd_id)` and read `.stdout`, `.stderr` and combined `.output`.
6. Poll completion state with `get_session_command(session_id, command_id)`, or inspect the whole session with `get_session(session_id)`.
7. Feed interactive input to a running command with `send_session_command_input(session_id, command_id, data)`.
8. For fully interactive terminal work, use the PTY API instead: create a PTY session with cols/rows plus cwd/envs, send keystrokes with `sendInput()`, adjust dimensions with `resize()`, and read the exit code on completion.
9. Always release resources with `delete_session(session_id)` (or the PTY kill call) — sessions are not garbage-collected for you.
10. Design log consumers to be resumable, since a dropped WebSocket during streaming requires re-attaching to the same cmd id. Official docs: https://www.daytona.io/docs/en/process-code-execution, https://www.daytona.io/docs/en/log-streaming, https://www.daytona.io/docs/en/pty