Add rendering performance example notebook

Port the `speed_up_illustration` notebook from spatialdata-notebooks into the
gallery as examples/performance.ipynb, adapted to repo conventions:

- Uses the cached `squidpy` visium_hne_sdata loader (its multiscale `hne`
  image drives the image sections) instead of a manually-downloaded zarr.
- Sections: automatic rasterization (single-scale images), automatic scale
  selection (multi-scale images), and the datashader backend for shapes/points.
- Slims the synthetic-data helper to build only the benchmarked element, and
  adds an untimed warmup so numba JIT / datashader setup isn't measured.
- Keeps the shape crossovers (~250 polygons/circles, ~225 multi-polygons) and
  shows that points are comparable across backends (no clean crossover).

Points benchmark starts at n=10: a single point has zero extent and crashes
the datashader backend (scverse/spatialdata-plot#724).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: timtreis

_static/img/performance.png

@@ -23,6 +23,15 @@ Draw regions of interest directly on a `spatialdata-plot` canvas with
 `sdata.pl.annotate(...)` and persist them as a `ShapesModel` element.
 :::
 
+:::{grid-item-card} Speeding up rendering
+:link: performance
+:link-type: doc
+:img-top: ../_static/img/performance.png
+
+Keep rendering fast on large data: automatic rasterization and scale
+selection for images, and the `datashader` backend for many shapes.
+:::
+
 ::::
 
 ```{toctree}
@@ -31,4 +40,5 @@ Draw regions of interest directly on a `spatialdata-plot` canvas with
 
 visium_mouse_brain
 interactive_annotate
+performance
 ```