improvement: switch run_command to process.exec with cwd/env/timeout support

[hjpinheiro]

Summary

Replaces the session-based execution pattern with a single process.exec() call, adding support for working directory, environment variables, and timeout.

What changed

Before (session-based — overhead per command):

sandbox.process.create_session(session_id)
response = sandbox.process.execute_session_command(session_id, ...)
# ... then delete_session in finally block

After (single call — cleaner, more capable):

response = sandbox.process.exec(command, cwd=cwd, env=env_vars, timeout=timeout)

New parameters

Parameter	Type	Description
cwd	string (optional)	Working directory (e.g. /home/daytona/project)
env_vars	string/JSON (optional)	Environment variables as JSON object string
timeout	number (optional)	Execution timeout in seconds

Other improvements

• Added create_text_message(response.result) so agents see command output directly in chat (previously only JSON was returned)
• Added _parse_env_vars() helper supporting both JSON strings and dicts

⚠️ Breaking change — output format

The JSON response changes from separate stdout/stderr fields to a combined output field:

// Before:
{"exit_code": 0, "stdout": "...", "stderr": "..."}

// After:
{"exit_code": 0, "output": "..."}  // combined stdout + stderr

This is because process.exec returns combined output in result. If separate stdout/stderr is important, the session-based approach can be kept as an alternative.

Testing

• py_compile passes
• YAML validated with yaml.safe_load()
• process.exec signature verified: (command, cwd=None, env=None, timeout=None) -> ExecuteResponse