[moetemp]

What changed?

This PR adds a Workflow factory for each Python resource.

• The client extension's get_<resource> factory of Add independent Python client generation and explicit API targets temporalio/nexgen#344 takes the flattened fields and the runtime types, as in client.get_stream(id="s1", body_type=Turn). The Workflow package exports a factory with the same name, as in stream_service.get_stream(id="s1", body_type=Turn). When an operation get_<resource> already returns the resource, as get-sender of the defaults sample does, the resource gets no factory. nexgen refuses any other operation with the factory's name. A factory makes no call.
• @nexus.flatten-in-api param="ref" fields="id backend" keep=true on a resource lets the factory take fields of one constructor parameter, or the whole value. So id="s1" and ref=ref both work.

TypeScript and .NET get the resource flatten on their factories in a follow-up PR.

Why?

The proposal builds a Stream with no call, like get_workflow_handle, from an id and a backend or from a whole ref. The ref is data, and code turns it into a Stream explicitly, so no converter turns a payload into a resource.

How did you test it?

Link to a test plan if any -

• Unit Tests
• Staging
• End to End Tests

Python tests build Streams with both factories, from the fields of the flattened parameter and from the whole value.

[moetemp]

Replaced by temporalio#373.