fix(ai): simplify resume-stream status flash fix

[gr2m]

Background

PR #12102 fixed a bug where useChat with resume: true would briefly flash status to submitted on page load when there is no active stream to resume (ready → submitted → ready).

However, the fix was more complex than necessary: it moved reconnectToStream out of the existing try-finally block, introduced a separate try-catch, a resumeStream closure variable with a non-null assertion (resumeStream!), and changed onFinish behavior (no longer called when there's nothing to resume).

Summary

This PR replaces the #12102 implementation with a simpler approach (net +3 lines changed in chat.ts vs +28/-13):

1. Skip setStatus('submitted') for resume-stream triggers — defer it until after reconnectToStream confirms there's an active stream
2. Set submitted only after reconnect succeeds — inside the existing try block, right after stream = reconnect

This keeps reconnectToStream inside the existing try-finally block, so:

• Errors are handled by the existing catch block (no duplicate try-catch)
• onFinish still fires in all cases (no behavior change vs pre-fix(ai): avoid status flash to 'submitted' during stream resumption with no active stream #12102)
• No resumeStream variable or non-null assertion needed

All tests from #12102 are preserved and pass.

Checklist

• Tests have been added / updated (for bug fixes / features)
• Documentation has been added / updated (for bug fixes / features)
• A patch changeset for relevant packages has been added (for bug fixes / features - run pnpm changeset in the project root)
• I have reviewed this pull request (self-review)

Related Issues

Simplifies #12102

[vercel]

When reconnectToStream returns null (no active stream to resume), the early return inside the try block triggers the finally block, causing a spurious onFinish callback with an empty/initial message state and unnecessary activeResponse creation/teardown.

[gr2m]

🤔 hm this might be legit

[nwalters512]

Responding here to #12102 (comment)

Candidly: I want the behavior change where onFinish is no longer called so I can remove the code I currently have to work around this unintuitive (IMO) behavior:

https://github.com/PrairieLearn/PrairieLearn/blob/2c180ff33a4220480de0efd791a153e8c85a9336/apps/prairielearn/src/ee/pages/instructorAiGenerateDraftEditor/components/AiQuestionGenerationChat.tsx#L552-L553

[gr2m]

I want the behavior change where onFinish is no longer called so I can remove the code I currently have to work around this unintuitive (IMO) behavior

Sounds good, and haven't seen any complaints, we should be good.