feature: add start_service and get_service_logs for background processes

[hjpinheiro]

Summary

Two new tools for running long-lived processes (web servers, daemons, build watchers) in sandboxes:

Tool	Purpose	SDK API
start_service	Start a background command asynchronously	process.create_session() + execute_session_command(var_async=True)
get_service_logs	Retrieve logs from a background service	process.get_session_command_logs()

Problem solved

run_command blocks until the command completes. This makes it impossible to start web servers, API servers, or any long-running process — the tool times out or blocks indefinitely.

start_service solves this by executing commands with var_async=True, which returns immediately with a session_id and cmd_id. The agent can then poll get_service_logs to check progress.

Design

start_service:

• Creates a session, executes command with var_async=True
• Returns session_id + cmd_id for log retrieval
• Parameters: sandbox_id (required), command (required)

get_service_logs:

• Retrieves current stdout/stderr from a running service
• Auto-discovers cmd_id from the session if not provided (fallback via get_session())
• Parameters: sandbox_id (required), session_id (required), cmd_id (optional)

Example workflow

1. create_sandbox → sandbox_id
2. start_service(sandbox_id, "python app.py") → session_id, cmd_id
3. get_service_logs(sandbox_id, session_id) → "Running on http://0.0.0.0:5000"
4. get_preview_url(sandbox_id, port=5000) → public URL

Testing

• Both .py files pass py_compile
• Both .yaml files pass yaml.safe_load
• provider/daytona.yaml validates with 9 total tools (7 original + 2 new)