First-party code execution sandbox for agent-generated code

Allow Trigger to run arbitrary, agent-generated code in a safe sandbox so agents can use a “run code” capability without a third-party service. Today, task containers are for running your own code with isolation and resource limits, not for executing untrusted code at runtime. This would add a dedicated execution path (e.g. a tool or task type) that accepts code + runtime, runs it in an isolated environment, and returns the result, so Trigger can cover the full agent + code-execution flow.