[Bug] OTEL metadata missing in `__init__` method of a workflow

[SF-300]

Hello,

What are you really trying to do?

I want to see proper OTEL metadata (spans and traces) propagated from the client invocation site all the way through workflows and activities, so I can visualize the full path in something like Datalust Seq.

Describe the bug

When a workflow starts an activity from its __init__ method, the OpenTelemetry context is not propagated. As a result, the activity runs with a missing/empty span context, even though the client call is wrapped in a valid span.

Minimal Reproduction

# /// script
# requires-python = ">=3.13"
# dependencies = [
#     "temporalio[opentelemetry]>=1.16.0",
# ]
# ///
import asyncio
from datetime import timedelta

from temporalio import activity, workflow
from temporalio.worker import Worker
from temporalio.contrib.opentelemetry import TracingInterceptor
from temporalio.testing import WorkflowEnvironment

from opentelemetry import trace
from opentelemetry.sdk.trace import TracerProvider
from opentelemetry.sdk.trace.export import ConsoleSpanExporter, SimpleSpanProcessor


# ---------- OpenTelemetry setup ----------
provider = TracerProvider()
provider.add_span_processor(SimpleSpanProcessor(ConsoleSpanExporter()))
trace.set_tracer_provider(provider)
tracer = trace.get_tracer(__name__)


@activity.defn
async def get_trace_id() -> int:
    return trace.get_current_span().get_span_context().trace_id


# ---------- Workflow ----------
@workflow.defn
class DemoWorkflow:
    @workflow.init
    def __init__(self) -> None:
        self._result = workflow.start_activity(
            get_trace_id,
            start_to_close_timeout=timedelta(seconds=5),
        )

    @workflow.run
    async def run(self) -> int:
        return await self._result


# ---------- Main ----------
async def main():
    async with await WorkflowEnvironment.start_local(
        interceptors=[TracingInterceptor()],
    ) as env:
        queue_name = "some-queue"
        async with Worker(
            env.client,
            task_queue=queue_name,
            workflows=[DemoWorkflow],
            activities=[get_trace_id],
        ):
            with tracer.start_as_current_span("invocation-site-span") as span:
                invocation_trace_id = span.get_span_context().trace_id
                activity_trace_id = await env.client.execute_workflow(
                    DemoWorkflow.run,
                    id="otel-update-workflow-id-2",
                    task_queue=queue_name,
                )
                assert activity_trace_id == invocation_trace_id


if __name__ == "__main__":
    asyncio.run(main())

Additional context

I think that WorkflowOutboundInterceptor should have a hook to wrap an __init__ of a workflow, but looking at how/where _instantiate_workflow_object is called, I don't think that it was planned for. So fixing this might require a refactoring of unclear scale to me.
If this issue is considered worth fixing, I'm willing to try fixing this and would greatly appreciate any clues (e.g. whether or not introducing another hook is the way to go here... or any other advice, really :)