PSK Reporter city lights: pre-scale the flat-map image off the main thread

Summary

Follow-up to #5479 (closes #5478). On the flat map, CityLightsItem::projPaint draws the full rendered city-lights image (up to 4096×4096, Format_ARGB32_Premultiplied) into the viewport with SmoothPixmapTransform on every repaint. The twilight/gamma/warmth render already runs on a worker, but this per-frame resample happens on the main thread.

Observed

While reviewing #5479 under software GL (xvfb / llvmpipe), enabling the layer produced two MainThreadWatchdog stalls, ~1.4 s and ~0.9 s, coinciding with the first two presentations of the image (initial load, then the twilight re-render). Not yet measured on a hardware GPU; it may be far cheaper there, but the work is still proportional to the source image size rather than the viewport.

Proposed change

After each CityLightsSource render completes, produce a viewport-resolution copy on the same QtConcurrent worker (scale to the item's current on-screen size for the visible world copies, respecting the device pixel ratio), and have CityLightsItem blit that instead of resampling the full image. Re-scale on zoom changes only, with the same debounce the source already uses for view changes. The full-resolution image stays as the source so the globe (which samples the original on the GPU) is unaffected.

Acceptance

• No watchdog stall when enabling the layer or when the twilight re-render lands, on both software GL and a hardware GPU.
• cityLightsFlatWrapAndOpacity in weather_radar_loading_test still passes (world wrapping and opacity unchanged).
• No visible change in output at the same zoom.

Related, not in scope here

Two other main-thread costs on the map worth their own issues if they show up in profiling: marker/path batch construction for large PSK Reporter result sets, and base-map tile decoding in QGeoView on reply arrival.

Start by tracing CityLightsItem::projPaint and the CityLightsSource render completion path, including the existing QtConcurrent worker and view-change debounce. Check how the current on-screen size and device pixel ratio are available, then run cityLightsFlatWrapAndOpacity in weather_radar_loading_test. Done means the flat map uses a worker-produced viewport-resolution image, zoom changes trigger rescaling, wrapping and opacity remain unchanged, and the watchdog stalls are absent on software and hardware GPU paths.