[BUG]: `layout.uirevision` doesn't play well with map subplot auto-fit

Description

When a map subplot uses fitbounds, layout.uirevision fails to preserve the user's subsequent pan/zoom across Plotly.react calls. The map snaps back to the auto-fit view even when the uirevision value is unchanged.

Screenshots/Video

https://github.com/user-attachments/assets/48f4da34-d74f-4958-a874-2bd9a3844181

Steps to reproduce

• Be on v4.0 (or the default branch after it gets released)
• Plot a scattermap with uirevision set and no explicit center/zoom:

  Plotly.newPlot(gd, [{
    type: 'scattermap',
    lon: [-75, -74, -73],
    lat: [40, 41, 42],
    mode: 'markers'
  }], {
    uirevision: 'v1',
    width: 400,
    height: 400
  });
  

• Pan the map with the mouse
• Call Plotly.react with the same uirevision:

  Plotly.react(gd, gd.data, { uirevision: 'v1', width: 400, height: 400 });
  

• Note that the map snaps back to the auto-fit view instead of keeping the panned view

Notes

• Discovered while working on #7884. The auto-fit and zoom-button behaviors landed there; this uirevision integration was deferred as out-of-scope.
• This works correctly for cartesian axes (via autorange) and for map subplots when the user has supplied an explicit center/zoom from the start. It only fails on the auto-fit path.

先在 layout.uirevision 不變的情況下重現 scattermap 案例，接著追蹤 Plotly.react 期間地圖子圖的自動適配路徑。將其與現有的明確 center/zoom 行為以及笛卡兒 autorange 行為進行比較。當 uirevision 不變時，使用者的平移或縮放能在多次 Plotly.react 呼叫之間保持不變，即視為完成。