Fix flaky Dev component polling tests

[gonzaloriestra]

WHY are these changes introduced?

Flaky tests: https://github.com/Shopify/cli/actions/runs/24343909340

FAIL   @shopify/app  src/cli/services/dev/ui/components/Dev.test.tsx > Dev > polls for preview mode
AssertionError: expected 1 to be greater than or equal to 2
 ❯ src/cli/services/dev/ui/components/Dev.test.tsx:511:58
    509|     expect(developerPreview.enable).toHaveBeenCalledTimes(1)
    510|     // fetchMode should be called multiple times due to polling
    511|     expect(developerPreview.fetchMode.mock.calls.length).toBeGreaterTh…
       |                                                          ^
    512| 
    513|     // unmount so that polling is cleared after every test

⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[1/2]⎯

 FAIL   @shopify/app  src/cli/services/dev/ui/components/Dev.test.tsx > Dev > doesn't poll for preview mode when the app does not support it
AssertionError: expected "spy" to not be called at all, but actually been called 4 times

Received: 

  1st spy call:

    Array []

  2nd spy call:

    Array []

  3rd spy call:

    Array []

  4th spy call:

    Array []



Number of calls: 4

 ❯ src/cli/services/dev/ui/components/Dev.test.tsx:570:44
    568|     `)
    569|     expect(developerPreview.enable).not.toHaveBeenCalled()
    570|     expect(developerPreview.fetchMode).not.toHaveBeenCalled()
       |                                            ^
    571| 
    572|     // Verify 'd' input doesn't trigger update when app doesn't suppor…

WHAT is this pull request doing?

Two timing-dependent test failures are fixed:

1. 'polls for preview mode': Replaced sleep-based timing with vi.waitFor() to reliably wait for the polling condition instead of hoping 50ms is enough.

2. 'doesn't poll for preview mode when the app does not support it': Used a locally-scoped mock to isolate from leaked polling intervals from previous tests that call the shared module-level developerPreview mock.

How to test your changes?

CI

Checklist

• I've considered possible cross-platform impacts (Mac, Linux, Windows)
• I've considered possible documentation changes
• I've considered analytics changes to measure impact
• The change is user-facing, so I've added a changelog entry with pnpm changeset add