Implement saving/loading MLEM objects (except for models and datasets)

[aguschin]

Would be useful if server and deployment objects were serializable/deserializable by MLEM, just like ML models and datasets are.

Use case: I define a MLEM Python object representing server or deployment, and I want to dump it to later use it from CLI. Such objects can be somewhat complex - see https://github.com/aguschin/art-expert-telegram-bot/blob/main/fastapi-deployment/serve_with_monitoring.py#L16 and it's counterpart, defined manually https://github.com/aguschin/art-expert-telegram-bot/blob/main/fastapi-deployment/server.mlem. Would be cool to simply call mlem.api.save(myserver, "server.mlem") and get the object by the last link dumped.