[Serve][2/n] Build/deploy flow for deployment-scoped actors

[abrarsheikh]

Adds plumbing for deployment_actors through the build and deploy pipeline. Deployment configs flow end-to-end, but deployment actors are not started yet (that’s in pr-2a-activation).

Changes

Application state

• deployment_to_deployment_actor_classes: extracts config-only actor class paths from the app config for the build task
• deployment_to_serialized_deployment_actors: merges serialized actor classes from code and config
• override_deployment_info(): injects _serialized_actor_class into deployment actor configs
• build_serve_application(): serializes code-defined actors and config-only actors, passes serialized_deployment_actors into get_deploy_args

Deploy utils

• get_deploy_args(): adds serialized_deployment_actors argument
• get_app_code_version(): uses deployment.deployment_actors from the schema and simplifies normalization

Deployment

• Deployment.options(deployment_actors=...) support

related #61464

[gemini-code-assist]

Code Review

This pull request adds the necessary plumbing for deployment-scoped actors, a significant new feature. The changes are well-structured across the application state, configuration, and utility modules. The new DeploymentActorConfig class is particularly well-designed, with robust serialization logic. I have one suggestion to improve the robustness of handling user-provided configurations.

_{Note: Security Review did not run due to the size of the PR.}

[gemini-code-assist]

It's good to check that cls_path is a string. It would be more robust to also check that name is a valid string here, as actor_cfg.get("name") could return None if the key is missing in the user-provided config. Using None as a dictionary key is likely not intended and could lead to issues downstream.

Suggested change

	if isinstance(cls_path, str):
	if isinstance(name, str) and isinstance(cls_path, str):

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, have a team admin enable autofix in the Cursor dashboard.}

[cursor]

json.dumps fails on non-dict DeploymentActorConfig objects

Low Severity

The simplified get_app_code_version now passes deployment.deployment_actors directly into json.dumps without normalization. The schema's deployment_actors field accepts Union[Dict, DeploymentActorConfig], but DeploymentActorConfig (a Pydantic BaseModel) is not natively JSON-serializable. The old code explicitly handled this by calling model_dump() and normalizing actor_class to a string representation. If a DeploymentActorConfig instance ends up in the list, json.dumps raises a TypeError.

 

[abrarsheikh]

in the context of this code block, deployment.deployment_actors will be dict