[moetemp]

What changed?

This PR lets a Python poll iterator yield the items of each page, with a point to resume from.

/// @nexus.poll python="next_read_request" items="records"
///   python-resume="resume_read_request" target="client"
read-records: func(request: read-records-request) -> read-records-response;

• items="<field>" names a list field of the response, and the iterator yields its items.
• python-resume= names an optional helper (page request, item) -> request that resumes after an item.
• The iterator is a PollIterator. Its next_request resumes after the last item it yielded, and it's None after the last page.

The stream sample's read yields each Record. The poll-service feed yields its executions in Workflow code and in a client.

It's based on #89, and it's on the fork for review before an upstream PR.

Why?

The proposal reads Records one by one, and a reader saves its place to resume after a restart. With the page iterator of temporalio#325, each caller looped over the Records and kept the offset itself.

How did you test it?

Link to a test plan if any -

• Unit Tests
• Staging
• End to End Tests

The poll-service Workflow reads the feed's items through real Nexus endpoints.