Add `Plugin` template support

[elliotgunton]

Is your feature request related to a problem? Please describe.
#1387

Plugin templates aren't directly supported by Hera; you have to use the model classes. It should be possible to use the plugins from the plugin directory using Hera's custom classes. Couple of examples:

Hello world

from hera.workflows import Workflow
from hera.workflows.models import Template, Plugin

class MyPlugin(Plugin):
    hello: dict = {}

w = Workflow(
    generate_name="hello-example-",
    entrypoint="main",
)

# unfortunately need to append as context manager doesn't
# work for model classes:

w.templates.append(
    Template(
        name="main",
        plugin=MyPlugin(
            hello={},
        ),
    )
)

For YAML:

apiVersion: argoproj.io/v1alpha1
kind: Workflow
metadata:
  generateName: hello-example-
spec:
  entrypoint: main
  templates:
  - name: main
    plugin:
      hello: {}

Slack plugin

from hera.workflows import Workflow
from hera.workflows.models import Template, Plugin

from pydantic.v1 import BaseModel


class Slack(BaseModel):
    text: str

class SlackPlugin(Plugin):
    slack: Slack

w = Workflow(
    generate_name="slack-example-",
    entrypoint="main",
)

w.templates.append(
    Template(
        name="main",
        plugin=SlackPlugin(
            slack=Slack(
                text="{{workflow.name}} finished!",
            ),
        ),
    )
)

For YAML:

apiVersion: argoproj.io/v1alpha1
kind: Workflow
metadata:
  generateName: slack-example-
spec:
  entrypoint: main
  templates:
  - name: main
    plugin:
      slack:
        text: '{{workflow.name}} finished!'

Describe the solution you'd like
These examples should be possible using a new PluginTemplate class which has the same behaviour under Workflow and DAG context managers as Scripts/Containers/etc.

Having the examples above in the repo is not currently possible as they fail testing due to the roundtrip test (assert workflow == HeraWorkflow.from_dict(workflow.to_dict())) not understanding how to interpret the value under plugin (i.e. it drops it). LHS vs RHS:

...
+ plugin=SlackPlugin(slack=Slack(text='{{workflow.name}} finished!')),
...

- plugin=Plugin(),

E.g. hello world would become:

from hera.workflows import PluginTemplate, Workflow
from hera.workflows.models import Template, Plugin


class HelloWorldPlugin(Plugin):
    hello: dict

with Workflow(
    generate_name="hello-example-",
    entrypoint="main",
) as w:
    PluginTemplate(
        name="main",
        plugin=HelloWorldPlugin(
            hello={},
        ),
    )

Using a DAG to show the callable behaviour:

from hera.workflows import DAG, PluginTemplate, Workflow
from hera.workflows.models import Plugin


class HelloWorldPlugin(Plugin):
    hello: dict

with Workflow(
    generate_name="hello-example-",
    entrypoint="main",
) as w:
    hello_template = PluginTemplate(
        name="hello",
        plugin=HelloWorldPlugin(
            hello={},
        ),
    )
    with DAG(name="main") as d:
        hello_template(name="hello-1")
        hello_template(name="hello-2")

Describe alternatives you've considered
Use the model classes, not much better than YAML 🙂

Additional context
Slack thread https://cloud-native.slack.com/archives/C03NRMD9KPY/p1747674689849999

[shuangkun]

Looking forward the feature！We use spark, ray plugins, etc. and hope to be able to write in Python.